LLM

Chat with Large Language Models (LLMs a.k.a. AI) using Textadept. Requires curl to be installed. This module can interact with OpenAI-compatible LLM servers, whether they are local (like mlx_lm) or remote. It can also interact with Ollama. Local LLM servers need to be running with one or more local models available.

Install this module by copying it into your ~/.textadept/modules/ directory or Textadept's modules/ directory, and then putting the following in your ~/.textadept/init.lua:

local llm = require('llm')
-- Remote, OpenAI-compatible server config.
llm.config.url = 'https://dev.example.com' -- if not OpenAI
llm.config.api_key = 'API_KEY'
-- Local mlx_lm server config.
llm.config.url = 'http://localhost:8080/v1'
-- Local Ollama config.
llm.config = llm.configs.ollama

Start a chat session from the "Tools > LLM (AI) > Chat..." menu.

Pressing Enter will prompt the model. Pressing Shift+Enter adds a new line without prompting the model. Typing @ will prompt you for an open file to inline as context to the model prompt.

If you have custom model options you want to use, like temperature and top_p, each server config has a models table with fields you can set. For example:

llm.config.model['mlx-community/Qwen3.5-9B-4bit'] = {
	stream = true, temperature = 0.7, top_p = 0.8, top_k = 20, max_tokens = 32768
}

The default model options enable streaming.

Note: if you are also using the scratch module, require this module after scratch, so that chats can be considered scratch buffers too.

llm.INDIC_LLM_END

The indicator number for where the LLM response ends.

llm.MARK_PROMPT

The marker number for prompt lines.

llm.MARK_PROMPT_COLOR

The color of prompt markers.

events.MODEL_RESPONSE

Emitted after a model is finished responding.

This could be used to provide a notification after a long wait time. Arguments:

• message: The model's entire response.

events.MODEL_RESPONSE_STREAM

Emitted after a model emits a streamed response.

Arguments:

• text: Partial model message.
• done: Whether or not this is the last part of the stream.

llm.chat([model[, system_prompt[, current_buffer]]])

Opens a new chat session with a model.

Parameters:

• model: String model name to chat with. If nil, the user is prompted for one.
• system_prompt: String system prompt to use for model. If both this and model are nil, the user has the option to specify a system prompt for the model.
• current_buffer: Whether or not to chat in the current buffer. The default value is false.

llm.config

The config table in configs to use.

Note: you may still have to configure things like the URL and API key. The default value is llm.configs.openai.

Fields:

• url: String URL and port the server is running on.
• models_endpoint: String REST endpoint that returns list of available models.
• model_name_key: String key whose value is the model name for each model in the REST response for models_endpoint.
• chat_endpoint: String REST endpoint for chatting with a model.
• chat_message: Function that accepts a REST response table fromchat_endpoint and returns its message object (not a string).
• done: Function that accepts a REST streaming response table from chat_endpoint and returns whether or not that endpoint is done streaming.
• curl_headers: Optional map of HTTP headers to send with curl requests to the server.
• api_key: Optional string API authorization key for the server. server does not support this option.
• model: Map of model names to maps of model-specific options like 'stream', 'think', 'temperature', 'top_p', etc.

Usage:

llm.config = llm.configs.ollama

llm.configs

Configurations for various LLM servers.

Fields:

• openai:
• ollama:

See also: llm.config

llm.prompt(input)

Prompts the current chat model with input.

A model's response will be printed when it is received.

Parameters:

• input: String input to prompt with. Any '@filename' references are replaced with their file's contents.

llm.prompts

Map of system prompt names (e.g. personas) to their prompt text. Users will typically select from this list prior to starting a chat session.

Usage:

llm.prompts.coding = 'You are a helpful coding assistant'

llm.show_system_prompt

Whether or not to show the system prompt when starting up a chat.

The default value is false.

llm.undo()

Undo the most recent chat message you submitted.

You will be able to edit and resend it.