# LLM — CLI Swiss Army Knife for Language Models

> Run prompts from the terminal, log everything to SQLite, manage 50+ providers via plugins. By Django co-creator. 11K+ stars.

## Install

Save as a script file and run:

# LLM — CLI Swiss Army Knife for Language Models

## Quick Use

```bash
pip install llm
```

```bash
# Set your API key
llm keys set openai
# Paste your key when prompted

# Run a prompt
llm "Explain quantum computing in one paragraph"

# Pipe content
cat error.log | llm "What's wrong here?"

# Start a conversation
llm chat -m claude-3.5-sonnet

# Use a different model
llm -m gpt-4o "Write a Python function to parse CSV"
```

Install plugins for more providers:
```bash
llm install llm-claude-3      # Anthropic Claude
llm install llm-gemini         # Google Gemini
llm install llm-ollama          # Local models
llm install llm-gpt4all        # Local GPT4All
```

---

## Intro

LLM is a command-line tool and Python library by Simon Willison (co-creator of Django) with 11,500+ GitHub stars for interacting with large language models. It works with OpenAI, Anthropic, Google, local models, and 50+ providers via a plugin ecosystem. The killer feature: every prompt and response is automatically logged to a local SQLite database, making it trivial to search, analyze, and audit your LLM usage. With prompt templates, embeddings support, and pipe-friendly design, LLM is the power user's tool for working with AI from the terminal.

Works with: OpenAI, Anthropic Claude, Google Gemini, Ollama, llama.cpp, GPT4All, 50+ providers via plugins. Best for developers who want a composable, scriptable interface to LLMs. Setup time: under 2 minutes.

---

## LLM Core Features

### Prompt Logging to SQLite

Every interaction is saved automatically:

```bash
# View recent prompts
llm logs

# Search logs
llm logs -q "quantum computing"

# SQL queries on your LLM history
llm logs --sql "SELECT model, count(*) FROM responses GROUP BY model"

# Export as JSON
llm logs --json | jq '.[] | .response'
```

Database stored at `~/.llm/logs.db` — open with any SQLite tool.

### Plugin Ecosystem

50+ plugins for different providers and features:

```bash
# Model providers
llm install llm-claude-3         # Anthropic Claude 3/4
llm install llm-gemini            # Google Gemini
llm install llm-ollama            # Local Ollama models
llm install llm-gpt4all          # Local GPT4All
llm install llm-mistral          # Mistral AI
llm install llm-replicate        # Replicate hosted models

# Features
llm install llm-cmd              # Generate shell commands
llm install llm-cluster          # Cluster embeddings
llm install llm-embed-jina       # Jina embeddings

# List installed plugins
llm plugins
```

### Prompt Templates

Save and reuse prompts:

```bash
# Create a template
llm --system "You are a code reviewer. Be concise." \
    "Review this code" --save review

# Use the template
cat main.py | llm -t review

# List templates
llm templates
```

### Embeddings

Generate, store, and search vector embeddings:

```bash
# Embed a collection of files
llm embed-multi docs -m 3-small --files docs/ '*.md'

# Semantic search
llm similar docs -c "How does authentication work?"

# Cluster documents
llm install llm-cluster
llm cluster docs 5  # 5 clusters
```

### Pipe-Friendly Design

```bash
# Summarize a file
cat README.md | llm "Summarize this in 3 bullets"

# Process command output
git diff | llm "Write a commit message for this diff"

# Chain with other tools
curl -s https://api.example.com/data | llm "Parse this JSON and list the top 5"

# Batch processing
cat urls.txt | while read url; do
  curl -s "$url" | llm "Extract the main topic" >> topics.txt
done
```

### Python Library

```python
import llm

model = llm.get_model("gpt-4o")
response = model.prompt("Explain transformers")
print(response.text())

# With conversation
conversation = model.conversation()
response1 = conversation.prompt("What is RAG?")
response2 = conversation.prompt("How do I implement it?")
```

---

## FAQ

**Q: What is LLM?**
A: LLM is a CLI tool and Python library with 11,500+ GitHub stars by Simon Willison for interacting with 50+ LLM providers, featuring automatic SQLite logging, prompt templates, embeddings, and a rich plugin ecosystem.

**Q: How is LLM different from other CLI AI tools?**
A: LLM's unique strength is its SQLite logging (every interaction is searchable/queryable) and plugin ecosystem (50+ providers). It's also the most Unix-philosophy-aligned AI tool — designed to pipe, compose, and script.

**Q: Is LLM free?**
A: Yes, open-source under Apache-2.0. You bring your own API keys.

---

## Source & Thanks

> Created by [Simon Willison](https://github.com/simonw). Licensed under Apache-2.0.
>
> [llm](https://github.com/simonw/llm) — ⭐ 11,500+

Thanks to Simon Willison for building the most composable CLI for language models.

---

<!-- ZH -->

## 快速使用

```bash
pip install llm
llm keys set openai
llm "用一段话解释量子计算"
```

---

## 简介

LLM 是 Django 联合创始人 Simon Willison 开发的命令行工具和 Python 库，拥有 11,500+ GitHub stars。支持 50+ LLM 提供商（通过插件），每次交互自动记录到本地 SQLite 数据库，支持提示模板和向量嵌入。终端用户与 LLM 交互的瑞士军刀。

适用于：OpenAI、Anthropic Claude、Gemini、Ollama、50+ 插件提供商。适合喜欢可组合、可脚本化 LLM 接口的开发者。

---

## 来源与感谢

> Created by [Simon Willison](https://github.com/simonw). Licensed under Apache-2.0.
>
> [llm](https://github.com/simonw/llm) — ⭐ 11,500+


---
Source: https://tokrepo.com/en/workflows/510fc429-b28e-4fd8-844b-b02881c84b03
Author: TokRepo精选