← 返回命令列表

Linux command

llm 命令

文本

复制后可按需替换文件名、目录或参数。

常用示例

Chat with default model

llm "[prompt]"

Use specific model

llm -m [gpt-4o] "[prompt]"

Interactive chat session

llm chat

Continue most recent conversation

llm -c "[follow up]"

List models

llm models

Pipe input

cat [file.txt] | llm "[summarize this]"

Use a system prompt

llm -s "[Reply as a pirate]" "[Hi there]"

Set an API key

llm keys set [openai]

Install a plugin

llm install [llm-claude-3]

说明

llm is a CLI and Python library for interacting with large language models. Out of the box it talks to OpenAI's API; through plugins it also supports Anthropic Claude, Google Gemini, Mistral, local Ollama, llama.cpp / GGUF, and many other providers and self-hosted models. Every prompt and response is logged to a local SQLite database (under ~/.config/io.datasette.llm/ on Linux or ~/Library/Application Support/io.datasette.llm/ on macOS) so previous conversations can be searched, exported, and replayed. The tool also supports prompt templates, system prompts, embeddings, similarity search, and pluggable backends.

参数

-m _MODEL_, --model _MODEL_
Model to use (e.g. gpt-4o, claude-3-5-sonnet, llama2).
-c, --continue
Continue the most recent conversation.
--cid _ID_
Continue a specific conversation by id.
-s _PROMPT_, --system _PROMPT_
Set a system prompt.
-t _NAME_, --template _NAME_
Use a named prompt template.
-o _KEY_ _VALUE_, --option _KEY_ _VALUE_
Pass a model-specific option (e.g. -o temperature 0).
-a _PATH_, --attachment _PATH_
Attach a file (image, PDF, audio) to the prompt for models that support attachments.
--no-stream
Disable token streaming and only print the final answer.
--key _KEY_
Use a specific API key for this call.

FAQ

What is the llm command used for?

llm is a CLI and Python library for interacting with large language models. Out of the box it talks to OpenAI's API; through plugins it also supports Anthropic Claude, Google Gemini, Mistral, local Ollama, llama.cpp / GGUF, and many other providers and self-hosted models. Every prompt and response is logged to a local SQLite database (under ~/.config/io.datasette.llm/ on Linux or ~/Library/Application Support/io.datasette.llm/ on macOS) so previous conversations can be searched, exported, and replayed. The tool also supports prompt templates, system prompts, embeddings, similarity search, and pluggable backends.

How do I run a basic llm example?

Run `llm "[prompt]"` in a terminal, then adjust file names, paths, flags, or remote targets for your system.

What does -m _MODEL_, --model _MODEL_ do in llm?

Model to use (e.g. gpt-4o, claude-3-5-sonnet, llama2).