Help for package rlmstudio

Title:

Access and Control LM Studio

Version:

0.3.0

Description:

Run local Large Language Models (LLMs) over many texts from 'R' without sending data to a third party. A community-maintained wrapper for the 'LM Studio' command line interface and API that provides functions to manage the local daemon and server, download and load models, and score, label, or generate text at scale.

License:

MIT + file LICENSE

URL:

https://jmgirard.github.io/rlmstudio/, https://github.com/jmgirard/rlmstudio

BugReports:

https://github.com/jmgirard/rlmstudio/issues

Depends:

R (≥ 4.1.0)

Imports:

cli, httr2, jsonlite, processx, utils

Suggests:

httptest2, httpuv, knitr, mockery, rmarkdown, testthat (≥ 3.3.0), withr

VignetteBuilder:

knitr

Config/testthat/edition:

Encoding:

UTF-8

Config/roxygen2/version:

8.1.0

NeedsCompilation:

Packaged:

2026-10-02 19:01:11 UTC; jmgirard

Author:

Jeffrey Girard

[aut, cph, cre]

Maintainer:

Jeffrey Girard <me@jmgirard.com>

Repository:

CRAN

Date/Publication:

2026-10-02 19:40:02 UTC

rlmstudio: Access and Control LM Studio

Description

Options

rlmstudio.quiet: A logical value. If TRUE, suppresses informational console messages and progress bars across the package. Defaults to FALSE. The quiet argument of list_models(), list_instances(), lms_chat_batch(), and lms_embed() defaults to NULL, which follows this option. quiet = TRUE hides the messages of that function or starts no progress bar, and quiet = FALSE prints the messages or starts the bar, also when this option is TRUE. Warnings show either way. The option does not change the quiet argument of lms_server_status(), which passes --quiet to the lms CLI.

Author(s)

Maintainer: Jeffrey Girard me@jmgirard.com (ORCID) [copyright holder]

Authors:

Jeffrey Girard me@jmgirard.com (ORCID) [copyright holder]

Check if the installed LM Studio CLI meets the minimum requirement

Description

Check if the installed LM Studio CLI meets the minimum requirement

Usage

check_lms_version(min_version = "0.4.0")

Arguments

min_version

Character string of the required version. Default is "0.4.0".

Value

A logical scalar: TRUE if the LM Studio CLI version meets or exceeds the specified min_version, and FALSE otherwise, including when lms_path() does not find the CLI.

Examples

## Not run: 
check_lms_version("0.4.0")

## End(Not run)

Check if LM Studio CLI is installed

Description

Uses the same lookup as lms_path(): the RLMSTUDIO_LMS_PATH environment variable, then the system PATH, then common installation directories.

Usage

has_lms()

Value

A logical scalar: TRUE if lms_path() finds the lms executable, and FALSE if it aborts.

Examples

## Not run: 
has_lms()

## End(Not run)

Help the user install or update LM Studio

Description

This function provides two methods for setting up LM Studio on your system. The "browser" method opens the official download page for the LM Studio desktop application (GUI). The "headless" method runs an automated installation script to install the llmster daemon and CLI, which is suitable for servers, containers, or users who prefer a GUI-less environment.

Usage

install_lmstudio(method = c("browser", "headless"))

Arguments

method

Character. Either "browser" (opens the GUI download page) or "headless" (installs the llmster daemon via script).

Details

If the headless installer exits with a status other than 0, the function aborts. The message gives the exit code and quotes the installer output, after "The installer said:". A byte that is not valid UTF-8 shows as ⁠<xx>⁠, its hex value. ANSI escape codes, such as color codes, cursor codes, and terminal links, are removed, and each run of whitespace becomes one space. A text longer than 1000 characters keeps at most its last 1000 characters, after "…". Any other error of the install step, such as a missing curl, aborts with "Headless installation failed." and the error message.

Value

Invisibly returns TRUE upon successful completion. This function is primarily utilized for its side effects of opening a web browser or executing system installation commands.

Examples

## Not run: 
# Open your default web browser to the download page
install_lmstudio(method = "browser")

# Attempt automatic headless installation via the command line
install_lmstudio(method = "headless")

## End(Not run)

List loaded model instances

Description

Retrieves the loaded model instances on the server via the LM Studio REST API, one row per instance, with the load configuration of each instance in columns. It is the view that ⁠lms ps⁠ prints, read from the same model list that list_models() reads, so it honors host and token.

Usage

list_instances(
  type = c("llm", "embedding"),
  quiet = NULL,
  host = "http://localhost:1234",
  token = NULL
)

Arguments

type

Character vector. The types of models to include. Defaults to c("llm", "embedding"). It must hold one or more elements, and no element can be NA, empty, or whitespace only. Each element must be valid in its declared encoding and not marked "bytes". Any other value, NULL and a factor included, aborts before the check for a running server. A type that no model has, such as "vlm", matches nothing. A classed filter matches by its value, and a class method such as as.character() does not change the match.

quiet

TRUE, FALSE, or NULL, the default. NULL follows the rlmstudio.quiet option. TRUE hides the message printed when no instance is found, and FALSE prints it, also when the option is TRUE. Any other value, NA included, aborts before the check for a running server. Does not suppress the abort raised when the server is not running.

host

Character. The host address of the local server.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

The model list does not carry four fields that ⁠lms ps --json⁠ reports: the generation status, the queued requests, the ttl, and the last-used time.

Value

A data.frame with one row per loaded instance of a model whose type is in type, with the columns that the "Columns" section describes. If there is no such instance, it returns a data.frame with zero rows and the four character columns, invisibly, and prints a message that quiet controls.

Columns

The first four columns are character columns:

id: the id of the instance.
key: the key of its model.
type: the type of its model, such as "llm" or "embedding".
display_name: the display name of its model, or NA when the model list gives none.

Rows follow the order of the models in the model list, then the order of the instances of a model. Two instances with the same id, or two models with the same key, give separate rows.

After the four columns, the frame has one column for each field name in the config object of an instance, in order of first appearance. When one config holds a name twice, the first value counts. A row whose config lacks the field, or has no config, holds NA there, or NULL in a list-column.

A column takes the field name as it is, with no name repair, so a name such as a-b needs backticks or [[. A field name that is empty, or that equals one of the four column names or the column name of an earlier field, takes the prefix config., again and again while the name still clashes. A field named id so gives the column config.id.

The column type follows the values of the field that are present and not null:

All strings give a character column.
All numbers give a double column, whole numbers included.
All booleans give a logical column.
No such values give a logical column of NA.

In these four kinds, a null value is NA. Any other mix, or any JSON object or array, gives a list-column. It holds each value as jsonlite::parse_json() returns it, and NULL where the field is absent or null.

Server not running

Functions that call the LM Studio REST API open a TCP connection to the hostname and port named in host before they send the request. A function that checks its own arguments does that first, so a bad model, job_id, input, inputs, messages, schema, ttl, previous_response_id, or batch_size, a bad type of list_models() or list_instances(), a bad TRUE or FALSE argument such as simplify, logprobs, quiet, force, or store, a store on the "openai" route of lms_chat() or lms_chat_batch(), a stream in the ... of a chat function, a logprobs in the ... of lms_chat_batch() or lms_chat_native(), a value in the ... of lms_chat_batch() that lms_chat() reads as an argument already given, an instructions or messages in the ... of lms_chat() or lms_chat_batch() on the route where lms_chat() sets it, or a name, id, or type string that is not valid in its encoding or is marked as bytes, aborts with an argument message and no condition class even when the server is down. A condition of class rlmstudio_no_server is raised when that connection cannot be opened. A refused connection raises it. So do an address the package cannot parse and a hostname that does not resolve. An address that neither accepts nor refuses the connection also raises it. That case waits for the operating system to give up, which can take a minute. Start the server with lms_server_start(), or give host the address that your server listens on.

The check reads the port and nothing else. Any process holding that port accepts the connection, so the condition is not raised even though no LM Studio server is there. The call then does not raise rlmstudio_no_server, and what it does depends on what answers. On a status-200 body that does not parse as JSON, the chat functions, lms_embed(), list_models(), list_instances(), lms_load(), lms_download(), lms_download_status(), and lms_unload_all() raise rlmstudio_bad_response. lms_unload() does not read the body, so it can report success. A body that parses as JSON but has another shape can come back unchanged with simplify = FALSE. With simplify = TRUE, the chat functions and lms_embed() raise rlmstudio_bad_response for it. list_models() and list_instances() raise it for a model list with another shape, and so do lms_unload_all() and lms_load() without force = TRUE, which read that list. lms_chat_openai() and lms_chat_openresponses() raise it for such a model list too, with either setting of simplify, through the model lookup that a reply from another model starts. lms_chat() raises it through them. lms_load(), lms_download(), and lms_download_status() raise it for a reply of their own with another shape, such as {}. A process that does not answer in HTTP gives an httr2_failure error. Use lms_server_ready() for the stronger test: it asks the host for a model list and reports TRUE only for a model list that list_models() can read.

lms_chat_batch() checks the server once before its first input, and lms_chat() checks it again for each input. If that check finds the server gone during the batch, the batch aborts with rlmstudio_no_server, and no request goes out after that. The condition then carries a results field, a list as long as inputs. Its elements before the lost input hold the values that format = "list" returns for those inputs. The element of the lost input and every element after it are NULL. The check before the first input adds no results field. A connection that fails after the check passes, such as a server that stops during a request, raises an httr2_failure error instead. That error aborts the batch and carries no results field.

lms_embed() checks the server before each request, and a request carries at most batch_size inputs. If a check after the first request finds the server gone, the call aborts with rlmstudio_no_server. Once a request has succeeded, the condition carries a results field. With simplify = TRUE, results is a matrix with NA in each row whose embedding did not arrive. With simplify = FALSE, it is a list with one element per batch, with NULL in the element of the request that ended the call and in every element after it.

API failure

A condition of class rlmstudio_api_error is raised when a REST call returns a response that the wrapper treats as a failure. The condition carries a status field, which holds the HTTP response status as an integer. It also carries a code field. The field holds the string at error.code of the response body, such as "model_not_found". It is NULL when the body does not parse, when error is not a JSON object, or when its code is not one string.

lms_chat_batch() aborts on it when its status is 401, 403, or 404, and when its status is 400 and its code is "model_not_found". No request goes out after that input. The condition then carries a results field that follows the rule for a lost server in the "Server not running" section: its elements before the failed input hold the values that format = "list" returns for those inputs, and the element of the failed input and every element after it are NULL. For any other status, the element of the failed input holds the condition, or NA where the result is text, and the batch warns once and goes on. See the details of lms_chat_batch().

lms_embed() follows the same rule for each request. A 401, 403, or 404 aborts the call, and once a request has succeeded the condition carries a results field, a matrix or a list as the "Server not running" section describes. Any other status fails the inputs of that request alone, and the call warns once and goes on. If every request fails, the call aborts with the first condition. See the details of lms_embed().

Malformed response

A condition of class rlmstudio_bad_response is raised when the server answers with a status the wrapper accepts and a body the wrapper cannot read. It is raised where a wrapper checks the body before it reshapes it, rather than indexing straight into whatever arrived. Eleven functions raise it for a body they cannot read: lms_embed(), lms_chat(), lms_chat_native(), lms_chat_openresponses(), lms_chat_openai(), list_models(), list_instances(), lms_load(), lms_download(), lms_download_status(), and lms_unload_all(). lms_chat() raises it through the chat function it calls. lms_unload_all() raises it through list_models(), and so does lms_load() unless force = TRUE. lms_chat_batch() raises it only as rlmstudio_model_mismatch, which the "Reply from another model" section of lms_chat_openai() describes.

All eleven raise it for a status-200 body that does not parse as JSON, such as an HTML page from a proxy, JSON text that stops part way, or an empty body. In the functions that take simplify, the body is parsed before simplify is read, so the condition is raised whatever simplify is. The body is parsed by its content and not by its Content-Type header, so valid JSON under text/plain is read as JSON. The body is read as JSON text and nothing else. A body whose text is a URL or the path of a file does not parse, and the package does not fetch the URL or read the file. The message says that the body did not parse as JSON and that something other than LM Studio may be answering on the host. It does not hold the body text.

The condition carries a status field, which holds the HTTP response status as an integer. Today the status is always 200: each of these functions reads the body only after a 200, and reports every other status as an rlmstudio_api_error instead.

Malformed model list

list_models() also raises it for a status-200 model list with the wrong shape. lms_unload_all() and lms_load() without force = TRUE raise it through list_models(). lms_chat_openai() and lms_chat_openresponses() apply these rules to the model list of their model lookup, as the "Reply from another model" section of lms_chat_openai() describes. A model list must follow four rules. Each field is read by its exact name, so a field named keyX does not stand in for key.

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

The rules are checked before the type and loaded filters, so an entry that the filters drop can still raise the condition. The message names the field or entry that broke a rule. lms_server_ready() applies the same rules and returns FALSE for a body that breaks one.

list_instances() applies the four rules too, and two more. They cover only an entry whose type is in its type argument and that has at least one loaded instance. The display_name of such an entry is a string, or absent, or null. The config of each of its instances is a JSON object, or absent, or null. list_models() and lms_server_ready() do not apply these two rules.

Examples

## Not run: 
lms_server_start()
lms_load("google/gemma-3-1b")

# One row per loaded instance, with its load configuration
list_instances()

# Only the loaded embedding models
list_instances(type = "embedding")

## End(Not run)

List available models

Description

Retrieves a list of models available on your system via the LM Studio REST API.

Usage

list_models(
  loaded = FALSE,
  type = c("llm", "embedding"),
  detailed = FALSE,
  quiet = NULL,
  host = "http://localhost:1234",
  token = NULL
)

Arguments

loaded

TRUE or FALSE. If TRUE, returns only currently loaded models. Defaults to FALSE. Any other value, NULL and NA included, aborts before the check for a running server.

type

detailed

TRUE or FALSE. Show all information about each model. Defaults to FALSE. Any other value, NULL and NA included, aborts before the check for a running server.

quiet

TRUE, FALSE, or NULL, the default. NULL follows the rlmstudio.quiet option. TRUE hides the message printed when no model is found, and FALSE prints it, also when the option is TRUE. Any other value, NA included, aborts before the check for a running server. Does not suppress the abort raised when the server is not running.

host

Character. The host address of the local server.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

A data.frame containing information about the available models. By default, it includes columns for state, type, display_name, key, architecture, and size_gb. If detailed = TRUE, it returns a comprehensive data.frame including all raw metadata columns provided by the API. Returns an empty data.frame if no models match the criteria.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Examples

## Not run: 
lms_server_start()
lms_download("google/gemma-3-1b")
lms_load("google/gemma-3-1b")

# List all downloaded models
list_models()

# List only currently loaded models
list_models(loaded = TRUE)

# Get detailed information about loaded text models
list_models(loaded = TRUE, type = "llm", detailed = TRUE)

## End(Not run)

Chat Completion with LM Studio

Description

Send a prompt to a locally running LM Studio model. This wrapper automatically routes your request to the appropriate subfunction based on the selected API type.

Usage

lms_chat(
  model,
  input,
  system_prompt = NULL,
  host = "http://localhost:1234",
  api_type = c("openresponses", "openai", "native"),
  logprobs = FALSE,
  simplify = TRUE,
  ...,
  schema = NULL,
  ttl = NULL,
  previous_response_id = NULL,
  store = NULL,
  token = NULL
)

Arguments

model

Character. The name of the loaded model. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

input

Character. The user prompt to send to the model, as one string. A character vector must hold no missing values, and its length must be one. Any other character value aborts before the check for a running server. To send several prompts, one request each, use lms_chat_batch(). A list is sent as given, as the input field or, on the "openai" route, as the content of the user message.

system_prompt

Character. An optional system prompt to guide model behavior.

host

Character. The base URL of the LM Studio server. Default is "http://localhost:1234".

api_type

Character. The LM Studio API endpoint to use. Options are "openresponses" (default), "openai", or "native".

logprobs

TRUE or FALSE. Whether to return the log probabilities of the generated tokens. Default is FALSE. Any other value, NULL and NA included, aborts before the check for a running server.

simplify

TRUE or FALSE. If TRUE, extracts the core text response. Default is TRUE. Any other value, NULL and NA included, aborts before the check for a running server.

...

Additional arguments passed to the selected API body. The package checks a stream here. A stream other than FALSE or NULL aborts before the call checks for a running server, because the package reads a whole reply and not a streamed one. An instructions here with api_type = "openresponses", the default, aborts before the request, because lms_chat() sets instructions from system_prompt on that route. A messages here with api_type = "openai" also aborts, because lms_chat() builds messages from system_prompt and input on that route. The message names the argument, and the abort has no condition class. Only these exact names abort. On the other routes, each goes into the request body.

schema

A JSON Schema that the reply must match, or NULL. It needs api_type = "openai", and any other api_type aborts before the request. See lms_chat_openai() for its form and for what is returned.

ttl

A whole number of seconds from 1 to .Machine$integer.max, or NULL. It is how long the model stays loaded with no request. It needs api_type = "openai", and any other api_type, the default included, aborts before the request. It has an effect only on a model that this request loads. The server loads a model that is not loaded yet when its just-in-time loading setting is on. A model that is already loaded keeps its idle time.

previous_response_id

One string, a value that carries a response_id attribute, or NULL. It names the stored reply that this chat continues. Pass the earlier reply itself, such as first after first <- lms_chat(...), or its id, attr(first, "response_id"). A value that carries a response_id attribute sends that attribute in place of the value, also when the value is a list or is itself an id. Only that exact attribute name is read. A value with no such attribute is sent as the id itself. So a reply that came back with no id, such as a native reply sent with store = FALSE, goes out as its own text, and the call raises rlmstudio_api_error with status 400. A reply text that breaks the id rules below, such as an empty text or a text of whitespace only, aborts before the request, as such an id does. The character vector that lms_chat_batch() returns with format = "vector" and the output column of its data frame carry no attribute. The data frame holds the ids in its response_id column.

The id, the attribute when there is one, must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the id is sent. NULL, the default, starts a new thread. It needs api_type = "native" or api_type = "openresponses". With api_type = "openai", an id or a value that carries one aborts before the request, because the OpenAI chat endpoint keeps no thread. As the id, NA, an empty string, a string of whitespace only, a value that is not a string, and more or fewer than one string abort before the check for a running server. For an attribute, the message names the attribute. Only this exact argument name is checked. A shortened name, such as previous, goes into the request body unchecked, under the name you wrote. An id that the server does not hold raises rlmstudio_api_error with status 400.

store

TRUE, FALSE, or NULL. Whether the server stores the reply, so that a later call can continue from its id. NULL, the default, sends no store field, so the server default applies, and that default stores the reply. TRUE and FALSE go out as a plain JSON true or false, also when the value has names, dimensions, or a class. Any other value, NA included, aborts before the check for a running server. A TRUE or FALSE needs api_type = "native" or api_type = "openresponses". NULL passes on every route. With api_type = "openai", TRUE or FALSE aborts before the request, because the OpenAI chat endpoint keeps no thread. Only this exact name is checked. A shortened name, such as sto, goes into the request body unchecked, under the name you wrote.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

This function calls lms_chat_openresponses(), lms_chat_openai(), or lms_chat_native(), according to api_type. It runs no request of its own. It can raise rlmstudio_no_server and rlmstudio_api_error through lms_chat_openresponses(), lms_chat_openai(), or lms_chat_native(). With simplify = TRUE, it can raise rlmstudio_bad_response through any of the three, for a reply that holds no readable answer text. With either setting of simplify, it raises rlmstudio_bad_response through any of the three for a status-200 body that does not parse as JSON. With either setting of simplify, it raises rlmstudio_model_mismatch through lms_chat_openresponses() or lms_chat_openai() for a reply from a model other than the one asked for.

Value

Depending on the arguments provided:

If simplify = FALSE, returns a parsed list of the raw JSON response.
If simplify = TRUE and logprobs = FALSE, returns a single character string containing the model's text response. With a schema, it returns the reply parsed into an R value instead.
If simplify = TRUE and logprobs = TRUE (and the chosen API type supports it), returns an object of class lms_chat_result containing both the text and a data.frame of token probabilities.

With simplify = TRUE and api_type = "native" or api_type = "openresponses", the value carries the id of the reply in a response_id attribute. Pass the value itself as previous_response_id to continue the thread, and the attribute is sent. The id is the response_id field of a native reply and the id field of an OpenResponses reply. The value has no attribute when that field is absent or is not one string. With api_type = "openai", or with simplify = FALSE, the value has no response_id attribute.

A native reply sent with store = FALSE has no response_id field, so its value has no attribute. An OpenResponses reply sent with store = FALSE still has an id, so its value carries the attribute, but the server does not hold that reply. A later call on the OpenResponses route that passes that id as previous_response_id raises rlmstudio_api_error with status 400 and the code "previous_response_not_found".

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed chat reply

lms_chat_native() and lms_chat_openresponses() raise it with simplify = TRUE when the reply holds no readable answer text. Both read the answer from the items of type "message" in the output array. They raise it when output is missing, empty, or not an array, or when an item in it is not a JSON object. They also raise it when no item has the type "message", as in a reply that holds only reasoning or a tool call. The text of a message must be one string. For lms_chat_native() that is the content of the item. For lms_chat_openresponses() it is the text of each part of type "output_text". For lms_chat_openresponses() only, the content of each message must be an array of JSON objects, and the messages together must hold at least one "output_text" part.

Apart from a body that does not parse as JSON, lms_chat_openai() raises it in four cases, all only with simplify = TRUE. The first case is a response whose choices field is missing, empty, or not an array, or whose first element is not a JSON object with a message object in it, so there is no reply to read. This case is raised with or without a schema, and with logprobs = TRUE as well. The second case is a reply that does not parse. A schema was given, logprobs = FALSE, and the reply content is not one string of valid JSON. The third case is reply content that is not one string, such as null, a missing content field, a number, or an array. A reply that holds only a tool call has null content. This case is raised without a schema, and with logprobs = TRUE with or without one. The fourth case is a cut-off reply. A schema was given, logprobs = FALSE, and the server reports the finish reason "length". This case is raised also when the reply content parses, because a reply that stops part way can parse to a wrong value, such as the first digit of a longer number. In the second, third, and fourth cases, if the server reports the finish reason "length", a length limit ended the reply. The limit is max_tokens or the context length of the model. The message then says so and names both.

With simplify = TRUE, lms_chat_native(), lms_chat_openresponses(), and lms_chat_openai() also raise it for a body that is a bare JSON value, such as 5, "s", or true. The message says that the response body is not a JSON object. A body of null gets the message about its missing output or choices field instead. lms_chat() can raise the condition through all three chat functions.

lms_chat_batch() does not abort on it, except on the subclass rlmstudio_model_mismatch, as the "Reply from another model" section of lms_chat_openai() says. The element of the failed input holds the condition, or NA where the result is text, and the batch warns once and goes on. See the details of lms_chat_batch().

A condition from lms_chat_openai() about its reply also carries two more fields. A condition from the model-list lookup does not, as the "Reply from another model" section of lms_chat_openai() says. The content field holds the reply content of the first choice, and the finish_reason field holds the finish reason of the first choice. Both are NULL for a response with no choices. In the third case, content holds the value that was read, which is NULL for null or missing content. For the second, third, and fourth cases, the message names the content field, so you can read what the model wrote without a second request. The other messages of the chat functions about the reply name simplify = FALSE, which returns the body unchanged, with one exception. A body that did not parse as JSON is checked before that argument is read, so its message points at the host instead. For such a body, the content and finish_reason fields of a condition from lms_chat_openai() are NULL. The messages of rlmstudio_model_mismatch and of the model-list lookup do not name simplify = FALSE, because the check runs with either setting of simplify.

Malformed logprobs

With simplify = TRUE and logprobs = TRUE, lms_chat_openresponses() also raises it for a logprobs value that breaks one of these rules. The logprobs value of each "output_text" part is checked. A logprobs value, token, logprob, or top_logprobs that is null or absent passes its rule. A null step or candidate breaks rule 2 or rule 6.

The value is an array.
Each step in the array is a JSON object.
The token of a step is a string.
The logprob of a step is a number.
The top_logprobs of a step is an array.
Each candidate in top_logprobs is a JSON object.
The token and logprob of each candidate follow rules 3 and 4.

The parts are checked in order, then the steps of a part, then the candidates of a step, one at a time. Within a step, rules 3, 4, and 5 come before the candidates. The message names the first broken rule that this order reaches. These checks run only after the text of every "output_text" part is read, so a reply that also has a bad text in any part gets the text message. Parts of other types, such as a refusal, are not checked, and with logprobs = FALSE no part is checked. Fields are read by their exact names, so a field whose name only starts with the one asked for, such as tokenX, reads as absent and gives NA in the data frame.

Cut-off reply

A warning of class rlmstudio_reply_cut_off is given when a length limit ended a reply that lms_chat_openai() returns as text. The server reports this with the finish reason "length". The limit is max_tokens or the context length of the model, and the reply does not say which one. The message names both. The warning is given with simplify = TRUE for reply content that is one string, in three settings: no schema, logprobs = TRUE, and a schema with logprobs = TRUE. The call returns what the same reply returns without the cut-off. A schema reply with logprobs = FALSE raises rlmstudio_bad_response instead, as the "Malformed chat reply" section says. The warning and that abort read the finish reason of the first choice, which is the choice that the call reads. No other element of choices is read. With simplify = FALSE, the call returns the body with every choice and no warning.

lms_chat() gives the warning through lms_chat_openai() with api_type = "openai". lms_chat_batch() gives one warning of this class for the whole batch in place of one for each input. It names the count and the positions of the cut-off inputs, and each of those elements keeps its reply. It comes after the warning about failed inputs. A batch that aborts with a results field gives this warning before the abort. It names the cut-off inputs whose replies the results field of the abort holds. The native and OpenResponses routes give no such warning.

The warning shows whatever quiet and the rlmstudio.quiet option say, because it is the only sign that an answer is not complete.

Reply from another model

lms_chat_openai() and lms_chat_openresponses() read the model field of each status-200 reply, with either setting of simplify. On LM Studio 0.4.25+1 with one chat model loaded, both routes answered a model name that the server could not find with status 200 and a reply from the loaded model. With two chat models loaded, they answered with status 400 and the code "model_not_found", as the "API failure" section describes. The model field names the loaded instance that answered. Its id can differ from the key that the call asked for.

If the model field equals the name that the call sent, the reply is accepted. If it differs, the call sends one request for the model list to api/v1/models on the same host, with the token of the call, and prints no message. The reply is accepted if a model whose key equals the asked name in any letter case has a loaded instance whose id is the model field. Otherwise the call aborts with a condition of class rlmstudio_model_mismatch. That condition is also an rlmstudio_bad_response, and its status is 200. Its model field holds the asked name, and its reply_model field holds the model field of the reply. The message names both. A condition from lms_chat_openai() also carries the content and finish_reason fields, both NULL.

The check runs before the reply is read, so it also runs with logprobs = TRUE, with a schema on lms_chat_openai(), and for a reply with no answer text. A reply is not checked if its body is not a JSON object, if it has no model field, or if its model is not one string or holds only whitespace. A body that is not a JSON object is returned with simplify = FALSE, and with simplify = TRUE it raises the error that the "Malformed chat reply" section describes. If the instance that answered is unloaded before the model-list request, the call aborts, also when that instance belongs to the asked model.

If the model-list request fails, the call raises the condition of that failure. The message opens with the label of the chat function, followed by "because the model-list lookup failed". A status other than 200 raises rlmstudio_api_error. A body that does not parse as JSON, or that breaks a rule of a model list in the "Malformed model list" section, raises rlmstudio_bad_response. A server that the port check before the request cannot reach raises rlmstudio_no_server. A condition from the lookup does not carry the content and finish_reason fields, also when lms_chat_openai() raises it.

lms_chat() runs the check through lms_chat_openai() and lms_chat_openresponses(). lms_chat_native() does not check the reply. On LM Studio 0.4.25+1, ⁠/api/v1/chat⁠ answered a name that it could not find with status 404, which raises rlmstudio_api_error.

lms_chat_batch() aborts at the first input that raises rlmstudio_model_mismatch, on the "openai" and "openresponses" routes. It also aborts at an rlmstudio_api_error with status 400 whose code is "model_not_found". No request goes out after that input. In both cases, the condition carries a results field that follows the rule for a lost server in the "Server not running" section.

Long prompts

A prompt longer than the context length of the loaded model fails. On LM Studio 0.4.25+1, with google/gemma-3-1b loaded at a context_length of 512, the ⁠/v1/responses⁠ and ⁠/api/v1/chat⁠ routes answered a longer prompt with status 500. The ⁠/v1/chat/completions⁠ route answered it with status 400. Each message began "The number of tokens to keep from the initial prompt is greater than the context length". The chat functions raise such a reply as an rlmstudio_api_error, and the message holds the server text. lms_chat_batch() fails that input alone and goes on to the next input. With format = "list", the element of that input holds the condition. With format = "vector", it holds NA.

To fit a longer prompt, load the model with a larger context_length in lms_load(). When the server reports it, the max_context_length column of list_models(detailed = TRUE) gives the largest context length that the model list reports for each model. If context_length is larger, the server loads the model with the asked value and gives no message. On LM Studio 0.4.25+1, it loaded google/gemma-3-1b at 65536 tokens, above its maximum of 32768. So lms_load() gives a warning of class rlmstudio_context_above_max that names both numbers, and then sends the load. The rlmstudio.quiet option does not hide it. The warning needs the model list, so it comes only with force = FALSE, and only when the list has a maximum for the model and the model is not loaded yet.

On LM Studio 0.4.25+1, the load endpoint answered a rope_frequency_scale field in ... with status 400 and the code "unrecognized_keys". So lms_load() cannot set RoPE scaling through that field. RoPE scaling stretches the position encoding of a model past its trained length.

To see how many tokens a prompt took, use lms_chat_batch(format = "data.frame"). Its input_tokens column holds the prompt token count that the server reports for each reply, on every route.

Batch Chat Completion with LM Studio

Description

Process a vector of inputs sequentially through LM Studio.

Usage

lms_chat_batch(
  model,
  inputs,
  system_prompt = NULL,
  format = c("vector", "list", "data.frame"),
  host = "http://localhost:1234",
  simplify = TRUE,
  quiet = NULL,
  ...,
  token = NULL
)

Arguments

model

Character. The loaded model name. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

inputs

Character vector. The prompts to process. Must hold at least one value and no missing values.

system_prompt

Character. Optional system prompt.

format

Character. Output format: "vector", "list", or "data.frame".

host

Character. Server URL.

simplify

TRUE or FALSE. If TRUE, parses outputs. Any other value, NULL and NA included, aborts before the check for a running server.

quiet

TRUE, FALSE, or NULL, the default. NULL follows the rlmstudio.quiet option. TRUE starts no progress bar, and FALSE starts one, also when the option is TRUE. cli draws a started bar only after a delay, two seconds by default. Any other value, NA included, aborts before the check for a running server. quiet does not hide the warnings about failed inputs, cut-off replies, a vector format that returns a list, or a logprobs = TRUE that the "native" route ignores.

...

Additional arguments passed to lms_chat, such as api_type, logprobs, schema, ttl, previous_response_id, or store. A schema, a ttl, a previous_response_id, a store, and the api_type that each needs are checked before the first call. A previous_response_id goes to every call, so each input continues the same stored reply. It can be an earlier reply of lms_chat() itself, and the response_id attribute of that reply is sent, by the rules of lms_chat(). A reply that came back with no id goes out as its own text, and each input then fails with status 400, unless that text breaks the id rules of lms_chat() and aborts first. The character vector of format = "vector" and the output column of format = "data.frame" carry no attribute, as described below. A store here goes to lms_chat() for every call, and it is checked there by the rules of lms_chat(). It must be TRUE, FALSE, or NULL, and NULL, the default, sends no store field. Any other value, NA included, aborts before the check for a running server. With api_type = "openai", TRUE or FALSE aborts before the check for a running server. Only this exact name is checked. A shortened name, such as sto, goes into each request body unchecked, under the name you wrote. With store = FALSE, a native reply has no response_id field. An OpenResponses reply still has an id, but the server does not hold that reply, so a later call on the OpenResponses route that passes it as previous_response_id gets status 400 with the code "previous_response_not_found". A logprobs here, or a shortened name that lms_chat() reads as logprobs, must be TRUE or FALSE. Any other value, NULL and NA included, aborts before the check for a running server. With api_type = "native", which has no logprobs, a logprobs of TRUE is ignored. The batch then returns what it returns with logprobs = FALSE, in each format. It gives one warning for the whole batch, after the check for a running server, even with quiet = TRUE. Two values here that lms_chat() reads as the same argument abort with a message that names the argument. Examples are ⁠logprobs = TRUE, logprobs = FALSE⁠, the shortened names ⁠log = TRUE, lo = FALSE⁠, and two previous_response_id values. An exact name and a shortened name, such as logprobs and log, do not abort, because R gives the shortened one to the ... of lms_chat(). An input here also aborts, because the batch passes each element of inputs as input. An input reaches ... only when inputs is given by its full name. Otherwise R reads input as a shortened inputs, so it becomes inputs, and the value given by position fills the next unnamed argument, such as system_prompt. These aborts come before every other check of ... and before the check for a running server. An instructions here on the "openresponses" route, the default, and a messages here on the "openai" route abort before the check for a running server, with the message that lms_chat() gives. A previous_response_id here follows the rules of lms_chat(). It must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the id is sent. The package checks a stream here. A stream other than FALSE or NULL aborts before the call checks for a running server, because the package reads a whole reply and not a streamed one.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

This function calls lms_chat() once for each element of inputs. It raises rlmstudio_no_server itself, before the first call.

An rlmstudio_bad_response that lms_chat() raises for one input fails that input alone, unless it is an rlmstudio_model_mismatch. So does an rlmstudio_api_error with any status other than 401, 403, or 404, unless its status is 400 and its code is "model_not_found". The batch goes on to the next input. Where the result is a list, or the output list-column that a schema gives, the element for that input holds the condition without its backtrace. An rlmstudio_bad_response for reply content that does not parse keeps that content in its content field. Where the result is text, the element holds NA. The result is text with format = "vector" when it returns a vector (simplify = TRUE, no schema, and logprobs = FALSE or the native route), and with a data frame whose replies are not parsed (no schema, or logprobs = TRUE). On the OpenResponses and OpenAI routes, the logprobs column holds NULL for a failed input. A reply with no readable answer text, such as one whose content is null, fails as an rlmstudio_bad_response in the same way. So does a status-200 body that does not parse as JSON, such as an HTML page from a proxy or an empty body. That input fails alone, and the other elements keep their replies. Use format = "list" to keep the conditions. The call gives one warning that names the count and the positions of the failed inputs. That warning shows even with quiet = TRUE.

An rlmstudio_api_error with status 401, 403, or 404 aborts the batch with that condition, and no request goes out after that input. Such a status comes from a fault that does not depend on the prompt, such as a token that the server refuses, so every later input fails in the same way. The condition carries a results field, as described in the "API failure" section below. No warning about failed inputs is given.

The batch aborts in the same way at an rlmstudio_api_error with status 400 and the code "model_not_found", and at an rlmstudio_model_mismatch. The "openai" and "openresponses" routes give these for a model name that the server cannot find, so every later input fails in the same way. See the "Reply from another model" section below.

A previous_response_id that the server does not hold gives status 400 with the code "invalid_value" on the native route and "previous_response_not_found" on the OpenResponses route. The batch does not abort at either one. Each such input fails alone, as described above, and the batch goes on to the next input.

An rlmstudio_no_server from lms_chat() also aborts the batch. Its results field holds the results so far, as described in the "Server not running" section below. An error of any other class aborts the batch unchanged.

Value

The return type depends on the format argument:

"vector": A character vector of responses, with NA for an input that failed. This format is only supported if simplify = TRUE, and logprobs = FALSE or api_type = "native". With a schema, it warns and returns the list instead.
"list": A list where each element is the response corresponding to the provided input, or the condition for an input that failed. With a schema, simplify = TRUE, and logprobs = FALSE, each element that did not fail is the parsed reply.
"data.frame": A data.frame containing input and output columns, with NA in output for an input that failed. If logprobs = TRUE on the "openresponses" or "openai" route, an additional list-column named logprobs is included, with NULL for an input that failed. The "native" route adds no logprobs column. With a schema and logprobs = FALSE, output is a list-column of parsed replies, with the condition in place of an input that failed. Columns read from each reply follow output, or logprobs when it is there, as described below. With an object schema and logprobs = FALSE, one column per schema property comes after them, as described below.

On the native and OpenResponses routes with simplify = TRUE, lms_chat() returns each reply that carries an id with a response_id attribute, as its help describes. With format = "list", each reply keeps it. So does each element of the list that format = "vector" returns with logprobs = TRUE on the OpenResponses route. The character vector of format = "vector" and the output column of format = "data.frame" carry no response_id attribute. The data frame holds the ids in its response_id column.

With api_type = "native" and format = "data.frame", the data frame ends with seven columns read from each reply: response_id, input_tokens, total_output_tokens, reasoning_output_tokens, tokens_per_second, time_to_first_token_seconds, and model_load_time_seconds. response_id is character, and it identifies the reply on the server. The other six are double, and they come from the stats object of the reply. A cell is NA when its field is absent or is not one value of the column type, a string for response_id and a number for the others. An empty string is a string, so an empty response_id is kept. If stats is absent or is not a JSON object, all six stats cells are NA. The server can leave a field out, such as model_load_time_seconds. Such a cell does not fail the input and gives no warning.

With api_type = "openresponses" or api_type = "openai" and format = "data.frame", the data frame has four columns read from each reply after output, or after logprobs when it is there. They are response_id, input_tokens, total_output_tokens, and reasoning_output_tokens. These are the first four native column names, but the servers send the values under other names:

response_id is the reply's id on both routes.
input_tokens is usage.input_tokens on the OpenResponses route and usage.prompt_tokens on the OpenAI route.
total_output_tokens is usage.output_tokens on the OpenResponses route and usage.completion_tokens on the OpenAI route.
reasoning_output_tokens is usage.output_tokens_details.reasoning_tokens on the OpenResponses route and usage.completion_tokens_details.reasoning_tokens on the OpenAI route.

The columns are there for every setting of logprobs and schema. response_id is character, and the three counts are double. The NA rule is the one for the native columns: a cell is NA when its field is absent or is not one value of the column type, and an empty string is kept. If usage is absent or is not a JSON object, all three count cells are NA. If the details object is absent or is not a JSON object, only reasoning_output_tokens is NA. Such a cell does not fail the input and gives no warning.

A reply with no readable answer text fails its input, whatever its other fields hold. The row of an input that failed holds NA in every column read from the reply. If every input failed, those columns are still there, response_id as character and the others as double. The vector and list formats add no such column.

With api_type = "openai", format = "data.frame", logprobs = FALSE, and an object schema, the data frame ends with one column per top-level property of the schema, after the four reply columns. An object schema has a type of "object", written as a string or as a list that holds that one string. Its properties is a list of one or more named entries. Each column has the property name, in the order of properties. The columns come from the schema and not from the replies, so they are there even when every input failed. The properties of a nested object get no columns of their own. Any other schema adds no column.

The type of a property column follows the type of the property. "string" gives character, "integer" gives integer, "number" gives double, and "boolean" gives logical. Each type can be a string or a list that holds that one string. A pair of one of these types and "null", such as c("integer", "null"), gives the same column type. Any other type, no type, or a property that is not a list gives a list-column.

A cell of a character, integer, double, or logical property column holds the field of the parsed reply when the field is one value of the column type. A one-item JSON array counts as one value, and an empty string is kept. A double column takes any number. An integer column takes an integer, or a whole number from -2147483647 to 2147483647. Otherwise the cell is NA, and the call gives no warning. The output column keeps the parsed reply, with the value that did not fit. A cell of a list-column holds the field as parsed, or NULL when the field is absent or null. A reply that is not a JSON object, such as an array or a number, gives NA and NULL cells. So does the row of an input that failed.

A property name that is empty, NA, repeated, or equal to another column name aborts before the call checks for a running server. The other column names are input, output, response_id, input_tokens, total_output_tokens, and reasoning_output_tokens. A data frame with logprobs = TRUE, and the list and vector formats, do not check the names.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed chat reply

Malformed logprobs

The value is an array.
Each step in the array is a JSON object.
The token of a step is a string.
The logprob of a step is a number.
The top_logprobs of a step is an array.
Each candidate in top_logprobs is a JSON object.
The token and logprob of each candidate follow rules 3 and 4.

Cut-off reply

The warning shows whatever quiet and the rlmstudio.quiet option say, because it is the only sign that an answer is not complete.

Reply from another model

Chat Completion via Native API

Description

Direct interface to LM Studio's v1 Native endpoint. Optimized for stateful chats and hardware control.

Usage

lms_chat_native(
  model,
  input,
  system_prompt = NULL,
  host = "http://localhost:1234",
  simplify = TRUE,
  ...,
  previous_response_id = NULL,
  store = NULL,
  token = NULL
)

Arguments

model

input

Character. The user prompt, as one string. A character vector must hold no missing values, and its length must be one. Any other character value aborts before the check for a running server. To send several prompts, one request each, use lms_chat_batch(). A list is sent as given in the input field.

system_prompt

Character. Optional system prompt.

host

Character. Server URL.

simplify

TRUE or FALSE. If TRUE, parses output to text. Any other value, NULL and NA included, aborts before the check for a running server.

...

Additional API arguments. This endpoint rejects a ttl field with status 400, which raises rlmstudio_api_error. The package checks a stream here. A stream other than FALSE or NULL aborts before the call checks for a running server, because the package reads a whole reply and not a streamed one. The package also checks each element named exactly logprobs. It must be TRUE, FALSE, or NULL, and any other value aborts before the check for a running server. The endpoint has no logprobs, so no logprobs field goes into the request body, and a TRUE warns.

previous_response_id

One string, a value that carries a response_id attribute, or NULL. It names the stored reply that this chat continues. Pass an earlier reply of this function itself, such as first, or its id, attr(first, "response_id"). A value that carries a response_id attribute sends that attribute in place of the value, also when the value is a list or is itself an id. Only that exact attribute name is read. A value with no such attribute is sent as the id itself. So a reply that came back with no id, such as a reply of this function sent with store = FALSE, goes out as its own text, and the call raises rlmstudio_api_error with status 400. A reply text that breaks the id rules below, such as an empty text or a text of whitespace only, aborts before the request, as such an id does. The character vector that lms_chat_batch() returns with format = "vector" and the output column of its data frame carry no attribute. The data frame holds the ids in its response_id column.

The id, the attribute when there is one, must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the id is sent. NULL, the default, starts a new thread. As the id, NA, an empty string, a string of whitespace only, a value that is not a string, and more or fewer than one string abort before the check for a running server. For an attribute, the message names the attribute. Only this exact argument name is checked. A shortened name, such as previous, goes into the request body unchecked, under the name you wrote. An id that the server does not hold raises rlmstudio_api_error with status 400 and the code "invalid_value". lms_chat() and lms_chat_batch() refuse an id or a value that carries one with api_type = "openai", because the OpenAI chat endpoint keeps no thread. lms_chat_openai() has no such argument. A previous_response_id in its ... goes into the request body unchecked, and the endpoint ignores it.

store

TRUE, FALSE, or NULL. Whether the server stores the reply, so that a later call can continue from its id. NULL, the default, sends no store field, so the server default applies, and that default stores the reply. TRUE and FALSE go out as a plain JSON true or false, also when the value has names, dimensions, or a class. Any other value, NA included, aborts before the check for a running server. Only this exact name is checked. A shortened name, such as sto, goes into the request body unchecked, under the name you wrote. lms_chat() and lms_chat_batch() refuse TRUE or FALSE with api_type = "openai".

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

If simplify = FALSE, returns a list representing the raw JSON response. A status-200 body that does not parse as JSON raises rlmstudio_bad_response with either setting of simplify. The body can hold a response_id for the reply and a stats object of token counts and timings. With api_type = "native" and format = "data.frame", lms_chat_batch() returns the id and six of the stats fields as columns. If simplify = TRUE, returns one character string: the content of every item of type "message" in the output array, pasted together in order with no separator. Items of other types, such as reasoning and tool calls, are skipped. A reply with no readable answer text raises rlmstudio_bad_response, as described below.

With simplify = TRUE, the string carries the response_id field of the reply in a response_id attribute. Pass the string itself as previous_response_id to continue the thread, and the attribute is sent. The string has no attribute when response_id is absent or is not one string. A reply sent with store = FALSE has no response_id field, so its string has no attribute, and there is no id to continue from. lms_chat_openresponses() reads the attribute from the id field of its reply instead. An OpenResponses reply sent with store = FALSE still has an id, so its value carries the attribute, but the server does not hold that reply. A later OpenResponses call that passes that id as previous_response_id raises rlmstudio_api_error with status 400 and the code "previous_response_not_found".

Server not running

API failure

Malformed response

Malformed chat reply

Chat Completion via OpenAI Compatibility API

Description

Direct interface to LM Studio's OpenAI-compatible endpoint. Uses the messages array format.

Usage

lms_chat_openai(
  model,
  messages,
  host = "http://localhost:1234",
  logprobs = FALSE,
  simplify = TRUE,
  ...,
  schema = NULL,
  ttl = NULL,
  token = NULL
)

Arguments

model

messages

The messages to send. Give an unnamed list with one element per message, such as list(list(role = "user", content = "Hi")), or a data frame with at least one row. A data frame is sent as one message per row, with one field per column. A cell that is NA is left out of its message. The call aborts before it checks for a running server if messages breaks one of these rules:

messages is a list or a data frame.
A data frame has at least one row, and any other list has at least one element.
A list that is not a data frame has no names. A single message not wrapped in a list, such as list(role = "user", content = "Hi"), breaks this rule.
Each element of such a list is a list of length one or more, and each of its fields has a name that is not NA and not empty. A data frame as an element breaks this rule.
A data frame has at least one column, and no column name is NA, empty, or repeated.
No column of a data frame has a row count that differs from the row count of the data frame that holds it. jsonlite cannot write such a column. The rule reads each data-frame column, and each atomic or list column with no class or with the class "AsIs" alone. It reads such columns in the data frame and in its data-frame columns at any depth. The row count of a data-frame column is its own row count. The row count of a column with a dim attribute, such as a matrix, is the first extent of the dim. The row count of any other column is its length. So a character column of length one in a two-row data frame breaks this rule. The rule does not read a column with another class, such as a Date, a factor, or a POSIXlt time. jsonlite writes a POSIXlt column of length one with its one value in each message, and the rule below reads that value in each row. A Date or factor column of a wrong length breaks the last rule, unless an earlier rule refuses the data frame first.
A data frame has no row in which every cell is NA or is a NULL cell of a list column. jsonlite leaves out an NA cell of another column and writes an NA or NULL list cell as null, so such a row holds no field value. A list() cell is sent as ⁠[]⁠ and a list(NA) cell as ⁠[null]⁠, so a row with such a cell is sent. A column with a dim attribute of any length whose first extent is the row count, such as a matrix or an array, counts as empty in a row when each of its cells in that row is empty. Such a column with no cells in a row, such as a 2-by-0 matrix, is not empty in that row, because jsonlite writes that row of it as ⁠[]⁠ or as nested empty arrays. A data-frame column counts as empty in a row when each of its own columns is empty in that row. So a data-frame column with no columns counts as empty in every row. A column that is neither an atomic vector nor a list, such as an environment, never counts as empty.
messages, when it is a list and not a data frame, has no dim attribute, such as a matrix or an array of messages.
A message that is a list has no dim attribute.
No list inside a message has a dim attribute. The rule reads each field, each list at any depth below a field, and each cell of a list column. jsonlite would send such a list as nested arrays with each cell in an array of its own. A list column of a data frame with one, three, or more dimensions also breaks this rule. An atomic matrix field is sent as an array of arrays. A list-matrix column of any data frame is sent, one row of cells for each message. jsonlite does not unbox a value inside a list-matrix cell, at any depth in a list, unless jsonlite::unbox() wraps it. A data frame in a cell is sent as an array of objects whose values are not boxed. So a length-one atomic cell is sent as a one-element array, such as ⁠["a"]⁠, and a NULL cell is sent as null. For example, a row with role "user", content "hi", and the cells "a", NULL, and list(k = "v") in a list-matrix column tags is sent as ⁠{"role":"user","content":"hi","tags":[["a"],null,{"k":["v"]}]}⁠.
No message, and no list or data frame inside a message or inside a cell or column of a data frame, has a name that is NA, empty, or repeated. jsonlite would send an NA or empty name as a number and rename a repeated name a to a.1. A list with no names passes. The names of a list column are not checked, because they are not sent. The names of an atomic vector and the column names of a matrix are not checked, because jsonlite sends those values as arrays.

The error for each rule above opens with "messages must be a data frame or an unnamed list of messages." and shows the form of a message. Four more rules are checked after them. Their error opens with "messages holds a field value that cannot be sent as JSON." and shows no form:

No function sits inside messages: not as a field, in a list below a field, as a column of a data frame, or in a cell of a list column. jsonlite would send a function as its source text. A function with a class gets the error of this rule, not the error of a later one.
Each value inside messages, in the places the rule above reads, is an atomic vector, a list, or NULL. An environment, a symbol, a call, a formula, an expression vector, an external pointer, or an S4 object breaks this rule, and the error names its type. The rule reads the storage type, so a class set by hand on such a value does not hide it. jsonlite would send such a value as printed text, as null, or as an object or array of other data, or it would fail. A NULL field passes and is sent as null. An S4 object whose class contains an atomic type, such as "numeric", passes this rule, and jsonlite sends its data part without its other slots. An S4 object whose class contains "list" passes, and the rule reads its elements. The rule does not read the class of a vector or a list. A vector with a class set by hand, such as structure(1L, class = "NULL"), passes, and jsonlite writes it by that class.
No number inside messages is NA, NaN, Inf, or -Inf. The rule reads a double or integer vector with no class or with the class "AsIs" alone, such as a field wrapped in I(). jsonlite would send such a number as the string "NA", "NaN", "Inf", or "-Inf". The rule reads a field, an element of a field, a list at any depth below a field, a cell of a list column or a list-matrix column, and a matrix or array column. An atomic column with no dim attribute, of a data frame at any depth, is the exception. jsonlite leaves an NA, NaN, Inf, or -Inf cell of such a column out of its message. There the rule refuses Inf and -Inf alone, and an NA or NaN cell is left out as a missing field. A number with another class, such as a Date, is written by its class, so as.Date(NA) is sent as null and as.Date(Inf) as "Inf". Below a message, the rule reads the parts of a list whose class is neither "AsIs" nor a data frame only when jsonlite writes that list as it writes the list without its class, such as a list with the class c("foo", "list"). It does not read the parts of a POSIXlt value.
jsonlite can write the value, with the options that the request uses. If it cannot, the error gives the jsonlite message. For a data frame, jsonlite then writes each top-level column alone, in column order, as a data frame with the same row count. If a column fails, the error names the first that fails. For a column d, the line reads ⁠Column "d" is the first column that jsonlite cannot write on its own.⁠ A list that is not a data frame gets the jsonlite message alone. This rule is checked last.

A list that is not a data frame is sent without its class attribute, and each of its messages is sent without its class attribute. A class on a field inside a message is kept, so a field wrapped in I() is sent as an array. A data frame and its columns keep their classes. A kept class that jsonlite has no method for, such as a field with the class "foo" alone, breaks the last rule.

The package checks the shape of messages. Of the field values, it refuses a function, a value that is not an atomic vector, a list, or NULL, a number with no class or with the class "AsIs" alone that jsonlite would send as a string or, in a data-frame column, would leave out because it is infinite, and a value that jsonlite cannot write. It does not check any other value of a role, the content, or another field of a message. The server checks them.

host

Character. Server URL.

logprobs

TRUE or FALSE. Whether to request logprobs (currently stubbed by LM Studio). Any other value, NULL and NA included, aborts before the check for a running server.

simplify

TRUE or FALSE. If TRUE, parses output to text. Any other value, NULL and NA included, aborts before the check for a running server.

...

Additional API arguments. A response_format here cannot be combined with schema. This function has no store argument, so a store here goes into the request body unchecked, whatever its value. The package checks a stream here. A stream other than FALSE or NULL aborts before the call checks for a running server, because the package reads a whole reply and not a streamed one.

schema

A JSON Schema, written as a named list, that the reply must match, or NULL for a free text reply. It is sent as the schema field of a response_format of type "json_schema", with the name "response" and strict set to true. A JSON array of one item must be written as a list, such as required = list("score"), or wrapped in I(). A plain vector of length one is sent as a single value, not as an array. An empty object nested in the schema, such as properties, is written setNames(list(), character()), because list() is sent as the empty array ⁠[]⁠. The package checks only that schema is a named list, an empty list, or NULL. The server checks the schema itself.

ttl

A whole number of seconds from 1 to .Machine$integer.max, or NULL to leave it out. It is how long the model stays loaded with no request. It has an effect only on a model that this request loads. The server loads a model that is not loaded yet when its just-in-time loading setting is on. A model that is already loaded keeps its idle time.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

If simplify = FALSE, returns a list representing the raw JSON response. A status-200 body that does not parse as JSON raises rlmstudio_bad_response with either setting of simplify. Otherwise, returns a character string containing the generated text. If logprobs = TRUE, it returns an lms_chat_result object with the log probabilities populated as NULL since they are currently stubbed in the LM Studio OpenAI endpoint. With simplify = TRUE, reply content that is not one string, such as the null content of a reply that holds only a tool call, raises rlmstudio_bad_response.

With a schema, simplify = TRUE, and logprobs = FALSE, the reply is parsed with jsonlite::parse_json(simplifyVector = TRUE) and the parsed value is returned. A JSON object becomes a named list, and an array of numbers becomes a vector. With simplify = FALSE or logprobs = TRUE, the reply stays a string. With a schema, simplify = TRUE, and logprobs = FALSE, a reply that a length limit ended raises rlmstudio_bad_response, also when it parses. With simplify = TRUE in any other setting, such a reply whose content is one string is returned with a warning. See the "Cut-off reply" section.

With simplify = TRUE, the reply is read from the first element of the choices field. No other element of choices is read. A request with n in ... therefore returns the first choice alone, and the cut-off warning and abort depend on the finish reason of that choice alone. With simplify = FALSE, the body holds every choice.

With either setting of simplify, a reply from a model other than the one asked for raises rlmstudio_model_mismatch. See the "Reply from another model" section.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed chat reply

Cut-off reply

The warning shows whatever quiet and the rlmstudio.quiet option say, because it is the only sign that an answer is not complete.

Reply from another model

Examples

## Not run: 
lms_chat_openai(
  model = "google/gemma-3-1b",
  messages = list(
    list(role = "user", content = "Rate 'Great value.' from 1 to 5.")
  ),
  schema = list(
    type = "object",
    properties = list(score = list(type = "integer")),
    required = list("score")
  )
)

## End(Not run)

Chat Completion via OpenResponses API

Description

Direct interface to LM Studio's OpenResponses endpoint. Supports logprobs and custom instructions.

Usage

lms_chat_openresponses(
  model,
  input,
  instructions = NULL,
  host = "http://localhost:1234",
  logprobs = FALSE,
  simplify = TRUE,
  ...,
  previous_response_id = NULL,
  store = NULL,
  token = NULL
)

Arguments

model

input

instructions

Character. Optional system instructions.

host

Character. Server URL.

logprobs

TRUE or FALSE. Whether to return token probabilities. Any other value, NULL and NA included, aborts before the check for a running server.

simplify

TRUE or FALSE. If TRUE, parses output to text and dataframe. If FALSE, returns raw list. Any other value, NULL and NA included, aborts before the check for a running server.

...

Additional API arguments (e.g., top_logprobs, temperature). This endpoint accepts a ttl field and ignores it. The model keeps the idle time that the server sets. The package checks a stream here. A stream other than FALSE or NULL aborts before the call checks for a running server, because the package reads a whole reply and not a streamed one.

previous_response_id

One string, a value that carries a response_id attribute, or NULL. It names the stored reply that this chat continues. Pass an earlier reply of this function or of lms_chat_native() itself, such as first, or its id, attr(first, "response_id"). A value that carries a response_id attribute sends that attribute in place of the value, also when the value is a list or is itself an id. Only that exact attribute name is read. A value with no such attribute is sent as the id itself. So a reply that came back with no id, such as a native reply sent with store = FALSE, goes out as its own text, and the call raises rlmstudio_api_error with status 400. A reply text that breaks the id rules below, such as an empty text or a text of whitespace only, aborts before the request, as such an id does. The character vector that lms_chat_batch() returns with format = "vector" and the output column of its data frame carry no attribute. The data frame holds the ids in its response_id column.

The id, the attribute when there is one, must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the id is sent. NULL, the default, starts a new thread. As the id, NA, an empty string, a string of whitespace only, a value that is not a string, and more or fewer than one string abort before the check for a running server. For an attribute, the message names the attribute. Only this exact argument name is checked. A shortened name, such as previous, goes into the request body unchecked, under the name you wrote. An id that the server does not hold raises rlmstudio_api_error with status 400 and the code "previous_response_not_found". lms_chat() and lms_chat_batch() refuse an id or a value that carries one with api_type = "openai", because the OpenAI chat endpoint keeps no thread. lms_chat_openai() has no such argument. A previous_response_id in its ... goes into the request body unchecked, and the endpoint ignores it.

store

TRUE, FALSE, or NULL. Whether the server stores the reply, so that a later call can continue from its id. NULL, the default, sends no store field, so the server default applies, and that default stores the reply. TRUE and FALSE go out as a plain JSON true or false, also when the value has names, dimensions, or a class. Any other value, NA included, aborts before the check for a running server. Only this exact name is checked. A shortened name, such as sto, goes into the request body unchecked, under the name you wrote. lms_chat() and lms_chat_batch() refuse TRUE or FALSE with api_type = "openai".

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

If simplify = FALSE, returns a list representing the raw JSON response. A status-200 body that does not parse as JSON raises rlmstudio_bad_response with either setting of simplify. Otherwise, returns one character string: the text of every part of type "output_text" in the items of type "message", pasted together in order with no separator. Reasoning items, tool calls, and parts of other types, such as a refusal, are skipped. If logprobs = TRUE and at least one of those parts carries log probabilities, returns an object of class lms_chat_result with that text and a data frame of the probabilities of every such part, in order. If no part carries them, it returns the string. A reply with no readable answer text raises rlmstudio_bad_response, as described below.

The data frame has one row for each candidate in the top_logprobs of each step, or one row with NA candidates for a step with none. It has five columns. step_token and step_logprob hold the token and the log probability of the step. candidate_token and candidate_logprob hold those of the candidate. step is an integer that numbers the steps from 1 in reply order, across all the parts. Each row of a step has the number of that step, so two steps with the same token keep different numbers. Each field is read by its exact name. A field that is null or absent gives NA, and so does a field whose name only starts with the one asked for, such as tokenX. With logprobs = TRUE, the logprobs value of each "output_text" part must follow seven rules, which the "Malformed logprobs" section below lists. A value that breaks one raises rlmstudio_bad_response, and the message names the first broken rule in the order the section below gives. Parts of other types are not checked. With logprobs = FALSE, the value is not read, and the call returns the text.

With simplify = TRUE, the string or the lms_chat_result carries the id field of the reply in a response_id attribute. Pass the value itself as previous_response_id to continue the thread, and the attribute is sent. The value has no attribute when id is absent or is not one string. A reply sent with store = FALSE still has an id, so its value carries the attribute, but the server does not hold that reply. A later call of this function that passes that id as previous_response_id raises rlmstudio_api_error with status 400 and the code "previous_response_not_found". lms_chat_native() reads the attribute from the response_id field of its reply instead. A native reply sent with store = FALSE has no response_id field, so its value has no attribute.

With either setting of simplify, a reply from a model other than the one asked for raises rlmstudio_model_mismatch. See the "Reply from another model" section.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed chat reply

Malformed logprobs

The value is an array.
Each step in the array is a JSON object.
The token of a step is a string.
The logprob of a step is a number.
The top_logprobs of a step is an array.
Each candidate in top_logprobs is a JSON object.
The token and logprob of each candidate follow rules 3 and 4.

Reply from another model

Start the LM Studio headless daemon

Description

Launches the llmster daemon in the background via the CLI. This is required in headless environments (such as Linux servers) before loading models or starting the local server.

Usage

lms_daemon_start()

Details

If the CLI exits with a status other than 0, the function aborts. The message gives the exit code and quotes what the CLI wrote, after "The CLI said:". The quoted text is the stderr text, or the stdout text if stderr holds only whitespace and escape codes. A byte that is not valid UTF-8 shows as ⁠<xx>⁠, its hex value. ANSI escape codes, such as color codes, cursor codes, and terminal links, are removed, and each run of whitespace becomes one space. A text longer than 1000 characters keeps at most its last 1000 characters, after "…".

Value

Invisibly returns the CLI exit code, 0.

Desktop Users

On desktop operating systems (macOS and Windows), running this command may actually launch the LM Studio desktop application to act as the backend engine. If the GUI is already open, this function will simply detect the active instance and return successfully. While safe to use, desktop users generally do not need to call this function and can just open the application manually.

Examples

## Not run: 
lms_daemon_start()

## End(Not run)

Check the global status of LM Studio

Description

Displays the overall status of the LM Studio backend via the CLI, including loaded models and the server state. This function works regardless of whether the backend was started via the desktop GUI or the headless daemon.

Usage

lms_daemon_status()

Value

A character vector of the raw CLI output.

Examples

## Not run: 
lms_daemon_status()

## End(Not run)

Stop the LM Studio headless daemon

Description

Stops the llmster daemon via the CLI. Use this to clean up system resources when you are completely finished using LM Studio in headless mode.

Usage

lms_daemon_stop(force = FALSE)

Arguments

force

TRUE or FALSE. If TRUE, attempts to stop the local server before shutting down the daemon. The daemon cannot be stopped while the server is actively running. Defaults to FALSE. If no server is running, the message of lms_server_stop() says so. Any other value, NULL and NA included, aborts before the lms CLI runs.

Details

If the CLI exits with a status other than 0, the function reads what the CLI wrote. If the text says that the daemon is part of LM Studio, the function prints an info message and returns FALSE. If the text says that the daemon is not running, it prints an info message and returns TRUE. Letter case does not matter. Any other failure aborts. The message gives the exit code, quotes what the CLI wrote after "The CLI said:", and gives a hint about force = TRUE. The quoted text is the stderr text, or the stdout text if stderr holds only whitespace and escape codes. A byte that is not valid UTF-8 shows as ⁠<xx>⁠, its hex value. ANSI escape codes, such as color codes, cursor codes, and terminal links, are removed, and each run of whitespace becomes one space. A text longer than 1000 characters keeps at most its last 1000 characters, after "…".

Value

Invisibly returns TRUE if the daemon stopped or was not running, and FALSE if the LM Studio GUI manages it.

Desktop Users

If the daemon is currently being managed by the LM Studio desktop application, the CLI does not stop it. The CLI intentionally prevents programmatic shutdowns of the GUI to avoid disrupting visual sessions. This function then returns FALSE, and the daemon keeps running. In this scenario, you must close the desktop application manually.

Examples

## Not run: 
lms_daemon_stop(force = TRUE)

## End(Not run)

Download a model via REST API

Description

Download a model via REST API

Usage

lms_download(
  model,
  quantization = NULL,
  host = "http://localhost:1234",
  ...,
  token = NULL
)

Arguments

model

Character. The model to download. Accepts model catalog identifiers (e.g., "openai/gpt-oss-20b") and exact Hugging Face links. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

quantization

Character. Optional. Quantization level of the model to download (e.g., "Q4_K_M"). Only supported for Hugging Face links.

host

Character. The host address of the local server. Defaults to "http://localhost:1234".

...

Additional arguments passed to the request.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

A character string containing the download job_id, or "already_downloaded", invisibly, if the model is already downloaded. The call aborts with rlmstudio_bad_response when the reply's status is "failed". It also aborts when the status is not "already_downloaded" and the reply holds no job_id string with a character that is not whitespace. It also aborts when the reply is not a JSON object whose status is a string.

Server not running

API failure

Malformed response

Malformed load or download reply

lms_load(), lms_download(), and lms_download_status() also raise it for a status-200 reply of their own with the wrong shape. Each reply must follow the rule of its function. Each field is read by its exact name. The rules check the type of a field and not its value, with four exceptions. The status of a load reply must be "loaded". A download reply whose status is "already_downloaded" needs no job_id. A download reply whose status is "failed" always aborts. The job_id of any other download reply must hold a character that is not whitespace.

A reply of lms_load() is a JSON object whose status is the string "loaded". With echo_load_config = TRUE, its load_config is also a JSON object.
A reply of lms_download() is a JSON object whose status is a string other than "failed". If the status is not "already_downloaded", its job_id is a string with a character that is not whitespace. For a "failed" status, the message says that LM Studio reports that the download failed. It names the reply's job_id if that is a string with a character that is not whitespace.
A reply of lms_download_status() is a JSON object whose job_id and status are strings. Its total_size_bytes, downloaded_bytes, and bytes_per_second are each a number, or absent, or null.

For the other faults, the message names the field that broke the rule, or it says that the body is not a JSON object. It also says that something other than LM Studio may be answering on the host.

Examples

## Not run: 
lms_server_start()

# Download a model by its HuggingFace identifier
job_id <- lms_download("google/gemma-3-1b")

# Download with a specific quantization level
lms_download("google/gemma-3-1b", quantization = "4bit")

## End(Not run)

Get the status of a download job

Description

Get the status of a download job

Usage

lms_download_status(job_id, host = "http://localhost:1234", token = NULL)

Arguments

job_id

Character. The unique identifier for the download job. Must be one id, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the id is sent.

host

Character. The host address of the local server. Defaults to "http://localhost:1234".

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

An object of class lms_download_status containing the download status.

Server not running

API failure

Malformed response

Malformed load or download reply

A reply of lms_load() is a JSON object whose status is the string "loaded". With echo_load_config = TRUE, its load_config is also a JSON object.
A reply of lms_download() is a JSON object whose status is a string other than "failed". If the status is not "already_downloaded", its job_id is a string with a character that is not whitespace. For a "failed" status, the message says that LM Studio reports that the download failed. It names the reply's job_id if that is a string with a character that is not whitespace.
A reply of lms_download_status() is a JSON object whose job_id and status are strings. Its total_size_bytes, downloaded_bytes, and bytes_per_second are each a number, or absent, or null.

For the other faults, the message names the field that broke the rule, or it says that the body is not a JSON object. It also says that something other than LM Studio may be answering on the host.

Examples

## Not run: 
lms_server_start()

job_id <- lms_download("google/gemma-3-1b")
status <- lms_download_status(job_id)
print(status)

## End(Not run)

Turn Text into Embedding Vectors

Description

Sends one or more texts to an embedding model and returns the vector that the model produced for each one. The texts go out in batches of at most batch_size, one request per batch, in the order given.

Usage

lms_embed(
  model,
  input,
  host = "http://localhost:1234",
  simplify = TRUE,
  ...,
  ttl = NULL,
  token = NULL,
  batch_size = 100,
  quiet = NULL
)

Arguments

model

Character. The loaded embedding model name. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

input

Character. The texts to embed. A vector of length n returns n embeddings, in the order given. Must hold at least one value and no missing values.

host

Character. Server URL.

simplify

TRUE or FALSE. If TRUE, the default, returns a numeric matrix with one row per input. FALSE returns a list with one element per request: the parsed response body, unchanged, or the condition of a request that failed. Any other value, NULL and NA included, aborts before the check for a running server.

...

Additional fields for the request body. LM Studio ignores a field it does not recognize, and two OpenAI fields are worth naming for that reason: LM Studio ignores dimensions, so asking for a narrower vector has no effect. encoding_format = "base64" is untested against LM Studio: a server that honors it returns embeddings this function cannot read, and the default simplify = TRUE path then aborts.

ttl

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

batch_size

A whole number from 1 to .Machine$integer.max. The most texts that one request carries. The default is 100. A value at least as large as length(input) sends every text in one request.

quiet

TRUE, FALSE, or NULL, the default. NULL follows the rlmstudio.quiet option. TRUE starts no progress bar, and FALSE starts one, also when the option is TRUE. cli draws a started bar only after a delay, two seconds by default. A bar starts only when the call sends more than one request. Any other value, NA included, aborts before the check for a running server. quiet does not suppress the warning about failed inputs.

Details

Before each request after the first, the function checks again that the server is running.

A request that fails with an rlmstudio_bad_response, or with an rlmstudio_api_error whose status is not 401, 403, or 404, fails the inputs it carried alone. Their rows hold NA, or their element of the simplify = FALSE list holds the condition without its backtrace. The call goes on to the next request and then gives one warning that names the count and the positions of the failed inputs. That warning shows even with quiet = TRUE. If every request fails, the call aborts with the condition of the first failed request, and no warning is given.

Two faults end the call at once, and no request goes out after them: a server that the check before a request finds gone, and an rlmstudio_api_error with status 401, 403, or 404. With simplify = TRUE, a third fault does the same: a request whose embeddings have another number of dimensions than those of an earlier request. It aborts with rlmstudio_bad_response. With simplify = FALSE, the bodies do not go through the matrix checks, so bodies of two widths are returned with no abort. Each abort after a request that succeeded carries a results field. With simplify = TRUE, it is the matrix so far, with NA in each row whose embedding did not arrive. With simplify = FALSE, it is the list so far, with NULL in the element of the request that ended the call and in every element after it. An abort before any request succeeded carries no results field. An error of any other class aborts the call unchanged.

On LM Studio 0.4.25+1, the server embeds only the first tokens of each text, up to the context length of the loaded model instance. For a text longer than that, it still returns a vector, with no error or warning, and the vector covers only the start of the text. The usage field of the reply reported 0 tokens on that version, so the count does not show the cut. The loaded_instances column of list_models(detailed = TRUE) gives the context_length of each loaded instance. To embed all of a long text, split it into pieces before the call. Or unload the model with lms_unload() and load it again with a larger context_length through lms_load(), up to the max_context_length column of list_models(detailed = TRUE). For a model that is already loaded, lms_load() does not load it again.

Value

If simplify = FALSE, a list with one element per request, in request order. Each element is the parsed JSON body of that request, or the condition of a request that failed. A call with one request returns a list of one. Otherwise, a double matrix with one row per input text and one column per embedding dimension. The row at position i holds the embedding that the response reported for the input at position i, or NA for an input whose request failed. The matrix carries no row or column names.

Server not running

API failure

Malformed response

Malformed embeddings

lms_embed() raises it on an embeddings block it cannot trust. The vectors it returns are placed by the index that the response reports, so a block with a missing, repeated, or out-of-range index would otherwise pair a vector with the wrong text and give back a matrix that is silently wrong.

lms_embed() reads each request on its own. A bad body, of either kind that the "Malformed response" section and this section describe, fails the inputs of that request alone, and the call warns once and goes on. The call aborts with the condition only if every request fails, and then with the condition of the first. With simplify = TRUE, it also aborts with rlmstudio_bad_response for a request whose embeddings have another number of dimensions than those of an earlier request. That condition carries a results field, a matrix as the "Server not running" section describes. See the details of lms_embed().

The messages of lms_embed() name simplify = FALSE, which returns the body unchanged, with one exception. A body that did not parse as JSON is checked before that argument is read, so its message points at the host instead.

Examples

## Not run: 
vectors <- lms_embed(
  model = "text-embedding-nomic-embed-text-v1.5",
  input = c("the first document", "the second document")
)
dim(vectors)

## End(Not run)

Load a model via REST API

Description

Load a model via REST API

Usage

lms_load(
  model,
  context_length = NULL,
  eval_batch_size = NULL,
  flash_attention = NULL,
  num_experts = NULL,
  offload_kv_cache_to_gpu = NULL,
  echo_load_config = FALSE,
  force = FALSE,
  host = "http://localhost:1234",
  ...,
  token = NULL
)

Arguments

model

Character. Unique identifier for the model to load. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

context_length

Integer. Maximum number of tokens that the model will consider. With force = FALSE, a value above the maximum in the model list gives a warning. See the "Long prompts" section.

eval_batch_size

Integer. Number of input tokens to process together in a single batch during evaluation.

flash_attention

TRUE, FALSE, or NULL. Whether to optimize attention computation. NULL, the default, leaves the field out of the request. Any other value, NA and "true" included, aborts before the check for a running server.

num_experts

Integer. Number of experts to use during inference for MoE models.

offload_kv_cache_to_gpu

TRUE, FALSE, or NULL. Whether KV cache is offloaded to GPU memory. NULL, the default, leaves the field out of the request. Any other value, NA and "true" included, aborts before the check for a running server.

echo_load_config

TRUE or FALSE. If TRUE, echoes the final load configuration in the response. Any other value, NULL and NA included, aborts before the check for a running server.

force

TRUE or FALSE. If TRUE, bypasses the check for currently loaded models and requests a new instance from the server. Note that this does not overwrite or replace the existing model; it loads a second concurrent instance into VRAM. Defaults to FALSE. Any other value, NULL and NA included, aborts before the check for a running server.

host

Character. The host address of the local server. Defaults to "http://localhost:1234".

...

Additional arguments passed to the API request body (useful for future API parameters).

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

Invisibly returns a character string of the loaded model identifier upon success, also when the model was already loaded. It is a plain string, with no class, names, or S4 bit. If echo_load_config = TRUE and this call loads the model, it instead invisibly returns a list containing the model's detailed load configuration.

Long prompts

To see how many tokens a prompt took, use lms_chat_batch(format = "data.frame"). Its input_tokens column holds the prompt token count that the server reports for each reply, on every route.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed load or download reply

A reply of lms_load() is a JSON object whose status is the string "loaded". With echo_load_config = TRUE, its load_config is also a JSON object.
A reply of lms_download() is a JSON object whose status is a string other than "failed". If the status is not "already_downloaded", its job_id is a string with a character that is not whitespace. For a "failed" status, the message says that LM Studio reports that the download failed. It names the reply's job_id if that is a string with a character that is not whitespace.
A reply of lms_download_status() is a JSON object whose job_id and status are strings. Its total_size_bytes, downloaded_bytes, and bytes_per_second are each a number, or absent, or null.

For the other faults, the message names the field that broke the rule, or it says that the body is not a JSON object. It also says that something other than LM Studio may be answering on the host.

Examples

## Not run: 
lms_server_start()
lms_download("google/gemma-3-1b")

# Load a model with default settings
lms_load("google/gemma-3-1b")

# Load a model with custom context length and flash attention enabled
lms_load("google/gemma-3-1b", context_length = 8192, flash_attention = TRUE)

## End(Not run)

Get the absolute path to the LMS executable

Description

Locates the LM Studio CLI (lms) on your system. It checks the RLMSTUDIO_LMS_PATH environment variable first, then the system PATH, and finally common installation directories.

Usage

lms_path()

Value

A character string specifying the absolute file path to the LM Studio executable (lms) on the user's system.

Examples

## Not run: 
lms_path()

## End(Not run)

Calculate Expected Scores and Uncertainty from Logprobs

Description

Takes a logprobs dataframe (from an lms_chat_result) and calculates the weighted average score, normalized probabilities, and uncertainty metrics for the first step of the reply, where the rating is.

Usage

lms_score_expected(lp_df, scale = 1:5)

Arguments

lp_df

A dataframe of logprobs (e.g., x$logprobs).

scale

Numeric vector. The valid labels (e.g., 1:5).

Details

The function reads the rows of the first step only. If lp_df has a step column and its first value is not NA, these are the rows whose step equals that value, wherever they sit in the frame. A later step with the same token as the first step does not count. Otherwise, these are the first run of consecutive rows whose step_token is identical to that of the first row, with NA equal to NA. In that case, with no step column or a first step of NA, two adjacent steps with the same token count as one step. Both rules start from the first row, so keep the rows in reply order: a sorted frame can start with a later step.

Of those rows, the function keeps the candidates whose token, read as a number, is in scale. Candidates that give the same label, such as "3", " 3", and "3.0", are summed into one label. The probabilities of the labels are then scaled to sum to 1.

Value

A named list containing three numeric elements (expected_value, weighted_sd, entropy) and a data.frame named probabilities with columns label and prob. probabilities has one row per label, in the order in which each label first appears, and entropy is computed over these rows. Returns NULL if lp_df is NULL or has no rows. Aborts if no candidate of the first step is in scale.

Examples

# Create a sample logprobs dataframe representing a model's generation step
mock_logprobs <- data.frame(
  step_token = rep("4", 3),
  step_logprob = rep(0, 3),
  candidate_token = c("4", "5", "3"),
  candidate_logprob = c(-0.105, -2.302, -3.506),
  step = rep(1L, 3),
  stringsAsFactors = FALSE
)

# Calculate the expected score and uncertainty metrics
lms_score_expected(mock_logprobs, scale = 1:5)

Check whether a host answers as a usable LM Studio server

Description

Sends one GET request to the model list endpoint at host and reports whether the answer came from an LM Studio server that this package can use. The answer is TRUE only when the request returns HTTP status 200 and the response body is a model list that list_models() can read. The two functions apply the same shape rules, which the "Malformed model list" section of list_models() states. An empty list counts, because a fresh LM Studio install has no models downloaded yet and its server still works.

Usage

lms_server_ready(host = "http://localhost:1234", timeout = 2, token = NULL)

Arguments

host

Character. The base URL of the LM Studio server. Defaults to "http://localhost:1234".

timeout

Numeric. The number of seconds to wait for the request before giving up. Defaults to 2. A host that opens the port and never answers reports FALSE after this many seconds.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

Use this in place of a port check. Another process holding the port, a server that has not finished starting, and a server that rejects the token all answer the port and all report FALSE here.

Value

TRUE or FALSE, always one value. This function raises no condition of its own for a network failure, a refused connection, an unparsable body, or a failed status. All of those report FALSE. A token that is not one character string and not NULL still aborts, because that is a fault in the call rather than a fact about the server.

Call faults that abort

A fault in the call is not a fact about the server, so it aborts rather than reporting FALSE. These messages come from the packages underneath. The httr2 ones name httr2's own arguments, url for host and seconds for timeout. The curl one names no argument at all. Six such faults are named below.

A host of NULL. httr2 reports that url must be a single string, not NULL.
A host of more than one string. httr2 reports that url must be a single string, not a character vector.
A host that is a character NA. httr2 reports that url must be a single string, not a character NA.
A host that is one string but cannot be parsed as a URL. curl reports that it failed to parse the URL and names the reason. A host holding a space gives "Malformed input to a URL function". An empty host gives a reason that says the URL has no host. Its wording depends on the libcurl version, for example "No host part in the URL" or "No host present".
A timeout below one millisecond. httr2 reports that seconds must be greater than 1 ms.
A timeout that is not one number, such as a string or a vector of two. httr2 reports that seconds must be a number, and names either the value or its type.

Those six are not the whole list. Any host that is not one string aborts the same way, whatever the reason, and httr2 names either the value or its type. A host of 1 gives "not the number 1". A host of list("a") gives "not a list". A host of character(0) gives "not an empty character vector".

The token fault named above aborts the same way. It comes from this package rather than from httr2.

Examples

## Not run: 
lms_server_start()

if (lms_server_ready()) {
  list_models()
}

# A server on another port, with a shorter wait
lms_server_ready(host = "http://localhost:8080", timeout = 0.5)

## End(Not run)

Start the LM Studio local server

Description

Launches the LM Studio local server via the CLI, allowing you to interact with loaded models via HTTP API calls.

Usage

lms_server_start(
  port = NULL,
  cors = FALSE,
  wait = 10,
  host = NULL,
  token = NULL
)

Arguments

port

Numeric or NULL. Port to run the server on. It must be one whole number from 1 to 65535, given as a number and not as an array or a string. NULL, the default, lets LM Studio use the last used port.

cors

Logical. Enable CORS support for web application development. Must be TRUE or FALSE. Defaults to FALSE. Any other value, NULL and NA included, aborts before the lms CLI runs.

wait

Numeric. How many seconds to keep asking the REST API whether it is ready. Defaults to 10. With wait = 0 the function sends no readiness request and returns as soon as the CLI does. A wait that is not one number, zero or more, aborts before the CLI runs.

host

Character or NULL. The base URL to ask. This says where to look for the server that was started. It does not change where the CLI starts it, which only port does. NULL picks a host as described below.

token

Character or NULL. An API token for the readiness request, for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

The CLI returns before the REST API answers. By default this function then keeps asking the REST API whether it is ready, for about wait seconds, and returns once it answers. A script that calls the REST API on the next line therefore no longer reports a missing server on a healthy machine. The budget is not a hard cap. No new request starts once wait seconds have passed, but a request already in flight is allowed one second to finish. With no host and no port, the function first asks the CLI which port the server uses, and that read runs before the wait seconds start to count.

Value

Invisibly returns an integer representing the system exit code (0 for success).

Which host the wait asks

The host is picked in this order.

A host you give wins.
With no host and a port, the host is ⁠http://localhost:<port>⁠.
With neither, the port that lms_server_status(json = TRUE) reports is read, and the host is ⁠http://localhost:⁠ plus that port.

The request carries token, read as described under that argument.

Five faults in the call abort before the CLI runs, with any wait: a bad wait, port, cors, host, or token. A bad host is one that is not NULL and that the readiness request cannot be built from, such as a vector of two strings, NA, an empty string, or "localhost:1234", which lacks ⁠http://⁠. That message names host and quotes the reason httr2 or curl gave. A bad token is one that is not one character string and not NULL.

If the CLI refuses the start and exits with a status other than 0, the function aborts. The message gives the exit code and quotes what the CLI wrote, after "The CLI said:". The quoted text is the stderr text, or the stdout text if stderr holds only whitespace and escape codes. A byte that is not valid UTF-8 shows as ⁠<xx>⁠, its hex value. ANSI escape codes, such as color codes, cursor codes, and terminal links, are removed, and each run of whitespace becomes one space. A text longer than 1000 characters keeps at most its last 1000 characters, after "…".

A wait that runs out does not abort. The server was already started and that cannot be undone, so the function raises a warning and returns the CLI exit code. A call with no host and no port whose port read yields nothing raises its own warning and sends no readiness request. If the readiness check itself aborts during the wait, the function raises a third warning. It names the host, quotes the abort message, and the function returns the CLI exit code. None of the three warnings is silenced by the rlmstudio.quiet option.

Examples

## Not run: 
# Start server on the default port and wait for the REST API
lms_server_start()

# Start server on a custom port with CORS enabled
lms_server_start(port = 8080, cors = TRUE)

# Return as soon as the CLI does, without waiting
lms_server_start(wait = 0)

# Wait longer, and ask a host the rules above would not pick
lms_server_start(wait = 60, host = "http://127.0.0.1:1234")

## End(Not run)

Check the status of the LM Studio server

Description

Displays the current status of the LM Studio local server via the CLI, including whether it is running and its configuration.

Usage

lms_server_status(
  json = FALSE,
  verbose = FALSE,
  quiet = FALSE,
  log_level = NULL
)

Arguments

json

TRUE or FALSE. Output the status in machine-readable JSON format. Any other value, NULL and NA included, aborts before the lms CLI runs.

verbose

TRUE or FALSE. Enable detailed logging output. Any other value, NULL and NA included, aborts before the lms CLI runs.

quiet

TRUE or FALSE. TRUE passes --quiet to the lms CLI, which suppresses all logging output. The rlmstudio.quiet option does not change it. Any other value, NULL and NA included, aborts before the lms CLI runs.

log_level

Character. The level of logging to use (e.g., "info", "debug").

Details

You can only use one logging control flag at a time (verbose, quiet, or log_level).

Value

By default, returns a character vector containing the raw CLI output. If json = TRUE and the jsonlite package is available, it returns a parsed list or data.frame of the status configuration.

Examples

## Not run: 
lms_server_start()

# Get basic status string
lms_server_status()

# Get status as a parsed JSON data frame
lms_server_status(json = TRUE)

## End(Not run)

Stop the LM Studio local server

Description

Stops the currently running LM Studio local server via the CLI.

Usage

lms_server_stop()

Details

If the CLI exits with a status other than 0, the function reads what the CLI wrote. If the text says "not running", the function prints an info message and returns the exit code invisibly, with no abort. Letter case does not matter. With no server running, the CLI exits with status 1 and says so. lms_daemon_stop() with force = TRUE shows the same message.

Any other failure aborts. The message gives the exit code and quotes what the CLI wrote, after "The CLI said:". The quoted text is the stderr text, or the stdout text if stderr holds only whitespace and escape codes. A byte that is not valid UTF-8 shows as ⁠<xx>⁠, its hex value. ANSI escape codes, such as color codes, cursor codes, and terminal links, are removed, and each run of whitespace becomes one space. A text longer than 1000 characters keeps at most its last 1000 characters, after "…".

Value

Invisibly returns an integer representing the system exit code (0 for success).

Examples

## Not run: 
lms_server_start()
lms_server_stop()

## End(Not run)

Unload a model from memory via REST API

Description

Unload a model from memory via REST API

Usage

lms_unload(model, host = "http://localhost:1234", ..., token = NULL)

Arguments

model

Character. Unique identifier (instance_id) of the model instance to unload. Must be one name, given as a single string. The string must be valid in its declared encoding and not marked "bytes". A class, names, and the S4 bit are removed before the name is sent.

host

Character. The host address of the local server. Defaults to "http://localhost:1234".

...

Additional arguments passed to the API request body.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Value

Invisibly returns a character string representing the unloaded instance_id upon success. It is a plain string, with no class, names, or S4 bit.

Server not running

API failure

Note

If you have loaded multiple instances of the same model using force = TRUE in lms_load(), the server assigns them unique instance identifiers (e.g., "google/gemma-3-1b" and "google/gemma-3-1b:2"). Passing the base model name to lms_unload() will only unload the primary instance. To unload duplicate instances, you must provide their exact instance_id, or use lms_unload_all() to clear everything.

Examples

## Not run: 
lms_server_start()
lms_download("google/gemma-3-1b")
lms_load("google/gemma-3-1b")

# Unload a single specific model
lms_unload("google/gemma-3-1b")

## End(Not run)

Unload all models from memory

Description

Retrieves a list of all currently loaded models and unloads them one by one.

Usage

lms_unload_all(host = "http://localhost:1234", ..., token = NULL)

Arguments

host

Character. The host address of the local server. Defaults to "http://localhost:1234".

...

Additional arguments passed to the API request body for each unload request.

token

Character or NULL. An API token for a server that requires authentication. NULL reads the rlmstudio.token option and then the RLMSTUDIO_API_TOKEN environment variable. See rlmstudio_token.

Details

This function calls list_models() to find the loaded instances, then calls lms_unload() once for each one. It raises rlmstudio_no_server itself, before the first call. It can raise rlmstudio_api_error through list_models() and through lms_unload(). It can raise rlmstudio_bad_response through list_models().

Value

Invisibly returns a character vector of the instance_ids that were successfully unloaded. If no models were currently loaded, it invisibly returns NULL.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Examples

## Not run: 
lms_server_start()
lms_download("google/gemma-3-1b")
lms_load("google/gemma-3-1b")

# Unload all currently loaded models to clear VRAM
lms_unload_all()

## End(Not run)

Create a new LM Studio chat result

Description

Internal constructor to create a structured object for responses containing log probabilities.

Usage

new_lms_chat_result(text = character(), logprobs = data.frame())

Arguments

text

Character. The generated text response.

logprobs

Dataframe. The token-level probability data.

Value

An object of class lms_chat_result.

Print an LM Studio chat result

Description

Custom print method for responses that include log probabilities. Displays the text clearly and provides a summary of the metadata.

Usage

## S3 method for class 'lms_chat_result'
print(x, ...)

Arguments

x

An object of class lms_chat_result.

...

Additional arguments passed to print.

Value

Invisibly returns the input object x.

Print method for LM Studio download status

Description

Print method for LM Studio download status

Usage

## S3 method for class 'lms_download_status'
print(x, ...)

Arguments

x

An object of class lms_download_status.

...

Additional arguments passed to print.

Value

Invisibly returns the input object x.

Examples

## Not run: 
lms_server_start()

job_id <- lms_download("google/gemma-3-1b")
status <- lms_download_status(job_id)
print(status)

## End(Not run)

Conditions raised by rlmstudio

Description

The functions in this package that talk to the LM Studio REST API raise four condition classes of their own. One of them, rlmstudio_model_mismatch, is also an rlmstudio_bad_response, and the "Reply from another model" section describes it. Each one is raised through cli::cli_abort(), so each one is an R error that you can catch by class with base::tryCatch(). The chat functions on the OpenAI route also give one warning class of their own, described in the "Cut-off reply" section.

Server not running

API failure

Malformed response

Malformed model list

The body is a JSON object whose models field is an array. The array can be empty.
Each entry of models is a JSON object. Its type and key are strings, and its loaded_instances is an array.
The size_bytes of an entry is a number, or absent, or null. So is its max_context_length.
Each entry of loaded_instances is a JSON object whose id is a string with a character that is not whitespace.

Malformed load or download reply

A reply of lms_load() is a JSON object whose status is the string "loaded". With echo_load_config = TRUE, its load_config is also a JSON object.
A reply of lms_download() is a JSON object whose status is a string other than "failed". If the status is not "already_downloaded", its job_id is a string with a character that is not whitespace. For a "failed" status, the message says that LM Studio reports that the download failed. It names the reply's job_id if that is a string with a character that is not whitespace.
A reply of lms_download_status() is a JSON object whose job_id and status are strings. Its total_size_bytes, downloaded_bytes, and bytes_per_second are each a number, or absent, or null.

For the other faults, the message names the field that broke the rule, or it says that the body is not a JSON object. It also says that something other than LM Studio may be answering on the host.

Malformed embeddings

Malformed chat reply

Malformed logprobs

The value is an array.
Each step in the array is a JSON object.
The token of a step is a string.
The logprob of a step is a number.
The top_logprobs of a step is an array.
Each candidate in top_logprobs is a JSON object.
The token and logprob of each candidate follow rules 3 and 4.

Cut-off reply

The warning shows whatever quiet and the rlmstudio.quiet option say, because it is the only sign that an answer is not complete.

Reply from another model

Examples

## Not run: 
tryCatch(
  list_models(host = "http://localhost:9999"),
  rlmstudio_no_server = function(cnd) {
    message("The server is not running: ", conditionMessage(cnd))
  },
  rlmstudio_api_error = function(cnd) {
    message("The API call failed with status ", cnd$status)
  },
  rlmstudio_bad_response = function(cnd) {
    message("The response body could not be read: ", conditionMessage(cnd))
  }
)

## End(Not run)

Authenticating to an LM Studio server

Description

An LM Studio server can require an API token. Every function in this package that calls the LM Studio REST API takes a token argument and sends the value as a bearer token in the Authorization header. Create and manage the token in the LM Studio app.

Where the token comes from

A function reads three sources and uses the first one that holds a value.

The token argument of the function you call.
The rlmstudio.token option, set with base::options().
The RLMSTUDIO_API_TOKEN environment variable.

When none of the three holds a value, the request carries no Authorization header. A source that is NULL or an empty string counts as unset. A token argument of any other shape, such as a vector of two strings or a number, aborts rather than falling through to the next source.

Keeping the token out of your output

The header is set with httr2::req_auth_bearer_token(), which marks it as redacted. Printing a request shows ⁠<REDACTED>⁠ in place of the value, and this package never puts the value into the message of a failed call.

Two things this package does not control can still show the value. The message of a failed call repeats the text the server sent, so a server that echoes the token back puts it there. An R error also prints a backtrace, which repeats your own calling line, so a token you write as a literal appears in it. Read the token from an environment variable to keep it out of both.

When the server rejects the call

A response with HTTP status 401 or 403 aborts with a condition of class rlmstudio_api_error, described in rlmstudio-conditions. If the request carried no token, the message tells you to set RLMSTUDIO_API_TOKEN or to pass token. If the request carried a token, the message tells you that the server rejected it.

Examples

## Not run: 
# Per call.
list_models(token = "your-token")

# For the session.
options(rlmstudio.token = "your-token")
list_models()

# For the machine, set RLMSTUDIO_API_TOKEN in your .Renviron file.

## End(Not run)

Validate an LM Studio chat result

Description

Internal validator to ensure the integrity of lms_chat_result objects.

Usage

validate_lms_chat_result(x)

Arguments

x

An object to validate.

Value

The validated object.

Run code with the LM Studio daemon active

Description

Temporarily starts the LM Studio headless daemon, executes the provided R expression, and then gracefully shuts the daemon and any active servers down. This is ideal for automated scripts and pipelines.

Usage

with_lms_daemon(code)

Arguments

code

An R expression to execute while the daemon is running.

Value

The result of the evaluated code.

Desktop Users

Be cautious using this wrapper if you already have the LM Studio GUI open. While the setup phase (lms_daemon_start) will succeed, the teardown phase (lms_daemon_stop) does not stop the daemon, because the CLI prevents programmatic shutdowns of the graphical interface. The teardown prints an info message, and the daemon keeps running. This wrapper is best reserved for strictly headless environments or fully automated scripts.

Examples

## Not run: 
result <- with_lms_daemon({
  lms_load("llama-3.1-8b")
  lms_chat("llama-3.1-8b", input = "Hello world!")
})

## End(Not run)

Package {rlmstudio}

rlmstudio: Access and Control LM Studio

Description

Options

Author(s)

See Also

Check if the installed LM Studio CLI meets the minimum requirement

Description

Usage

Arguments

Value

Examples

Check if LM Studio CLI is installed

Description

Usage

Value

Examples

Help the user install or update LM Studio

Description

Usage

Arguments

Details

Value

Examples

List loaded model instances

Description

Usage

Arguments

Details

Value

Columns

Server not running

API failure

Malformed response

Malformed model list

See Also

Examples

List available models

Description

Usage

Arguments

Value

Server not running

API failure

Malformed response

Malformed model list

See Also

Examples

Chat Completion with LM Studio

Description

Usage

Arguments

Details

Value

Server not running

API failure

Malformed response

Malformed model list

Malformed chat reply

Malformed logprobs

Cut-off reply

Reply from another model

Long prompts

Batch Chat Completion with LM Studio

Description

Usage

Arguments

Details

Value

Server not running

API failure

Malformed response

Malformed model list

Malformed chat reply

Malformed logprobs

Cut-off reply

Reply from another model

Chat Completion via Native API

Description

Usage