# Docs

- Jan Desktop: The desktop app

  - **Get started**
  - [Overview](/desktop): Jan is a local, privacy-first AI — a Claude-class alternative you own and control
  - [QuickStart](/desktop/quickstart): Get started with Jan and start chatting with AI in minutes.
  - Install 👋 Jan
    - [Mac Installation](/desktop/install/mac): Download Jan on Mac to run AI models locally. Jan is a free, open-source ChatGPT alternative to run offline.
    - [Windows Installation](/desktop/install/windows): Download Jan on Windows to run AI models locally. Jan is a free, open-source ChatGPT alternative to run offline.
    - [Linux Installation](/desktop/install/linux): Download Jan on Linux to run AI models locally. Jan is a free, open-source ChatGPT alternative to run offline.
  - Models
    - [Jan-Code-4B](/desktop/jan-models/jan-code-4b): Lightweight 4B code-tuned model for fast local coding assistance and agentic workflows
    - [Jan-v3-4B](/desktop/jan-models/jan-v3-4b-base-instruct): 4B parameter instruct model distilled from a larger teacher, optimized as a fine-tuning base
    - [Jan-v2-VL](/desktop/jan-models/jan-v2-vl-med): 8B vision-language model for long-horizon agentic automation in real software environments
    - [Jan-v1](/desktop/jan-models/jan-v1): 4B parameter model with strong performance on reasoning benchmarks
    - [Jan-Nano-128k](/desktop/jan-models/jan-nano-128): Jan Models
    - [Jan Nano](/desktop/jan-models/jan-nano-32): Jan-Nano-Gguf Model
    - [Lucy](/desktop/jan-models/lucy): Compact 1.7B model optimized for web search with tool calling
  - Integrations
    - [Claude Code](/desktop/integrations/claude-code): Use Jan as a local AI backend for Claude Code
    - [Agents](/desktop/integrations/openclaw): Autonomous AI agents that run on your own hardware
    - [MCP Servers](/desktop/integrations/mcp-servers): Extend Jan with Model Context Protocol servers

  - **Core concepts**
  - [Cowork](/desktop/cowork): Use Cowork to plan tasks, work with files, and review the results in Jan.
  - [Projects](/desktop/projects): Organize your AI conversations with shared instructions and files
  - [Assistants](/desktop/assistants): A step-by-step guide on customizing and managing your assistants.
  - [Web Search](/desktop/web-search): Let models search and read the web using Jan's built-in web_search and web_fetch tools
  - [File Upload](/desktop/file-upload): Upload images and documents to chat with Jan
  - [Model Management](/desktop/manage-models): Manage your interaction with AI models locally.
  - [Local API Server](/desktop/api-server): Run Jan's OpenAI-compatible API server on your local machine for private, offline AI development.
  - [Settings](/desktop/settings): Explore how to adjust Jan's settings to suit your specific requirements.
  - Local AI Engine
    - [Local AI Engine (llama.cpp)](/desktop/local-engine/llama-cpp): Understand and configure Jan's local AI engine for running models on your hardware.
    - [MLX](/desktop/local-engine/mlx): Run AI models with MLX on Apple Silicon
  - [Model Settings](/desktop/model-parameters): Customize how your AI models behave and perform.

  - **Reference**
  - [API Reference](/desktop/api-preference): API endpoints supported by Jan's local API server
  - [Jan CLI and Desktop](/desktop/cli): Install Jan Agent separately from Jan Desktop and connect it to a local model through the Local API Server.
  - [Using the Model Context Protocol (MCP) in Jan](/desktop/mcp): Manage your interaction with AI locally.
  - [MCP routing telemetry](/desktop/mcp-routing-telemetry): Optional callbacks for smart MCP tool routing — latency, fallbacks, and router usage
  - Cloud Providers
    - [Anthropic](/desktop/remote-models/anthropic): Learn how to integrate Anthropic with Jan for enhanced functionality.
    - [Google](/desktop/remote-models/google): Learn how to integrate Google with Jan for enhanced functionality.
    - [Groq](/desktop/remote-models/groq): Learn how to integrate Groq API with Jan for enhanced functionality.
    - [Mistral AI](/desktop/remote-models/mistralai): A step-by-step guide on integrating Jan with Mistral AI.
    - [OpenAI](/desktop/remote-models/openai): A step-by-step guide on integrating Jan with OpenAI.
    - [Azure OpenAI](/desktop/remote-models/azure): A step-by-step guide on integrating Jan with Azure OpenAI.
    - [OpenRouter](/desktop/remote-models/openrouter): A step-by-step guide on integrating Jan with OpenRouter.
    - [Hugging Face](/desktop/remote-models/huggingface): Learn how to integrate Hugging Face models with Jan using the Router or Inference Endpoints.
    - [Custom Endpoints](/desktop/remote-models/custom-endpoint): Connect Jan to any OpenAI-compatible or Anthropic-compatible API endpoint, including self-hosted servers and proxies.
  - [Jan Data Folder](/desktop/data-folder): A guide to Jan's data structure.
  - [Troubleshooting](/desktop/troubleshooting): Explore solutions for common issues and optimize Jan's performance with this comprehensive troubleshooting guide.
  - [Privacy](/desktop/privacy): Jan is an app that allows you to own your AI. We prioritize your control over your data and explain what data we collect and why.
- Jan Agent: The standalone agent

  - **Get started**
  - [Overview](/agent): Jan Agent is a terminal AI agent that reads, writes, and runs code in your project - powered by the models you already configured.
  - [QuickStart](/agent/quickstart): Install Jan Agent, connect a provider, and finish your first task in the terminal.
  - [Using Jan Agent](/agent/console): A tour of the console - the screen, approving work, and what to reach for next.
  - [Slash Commands](/agent/slash-commands): Every slash command available in the Jan Agent console.
  - [Keybindings](/agent/keybindings): Keyboard shortcuts for the Jan Agent console.

  - **Working with the agent**
  - [Run Modes](/agent/run-modes): Default, safe, and plan - how much freedom the agent has in a session.
  - [Sessions](/agent/sessions): Saved conversations, resuming, rewinding, and queueing messages while the agent works.
  - [Goals and Todos](/agent/goals-and-todos): Give the agent a finish line, and track multi-step work as it goes.
  - [Context and Compaction](/agent/context): How Jan Agent manages a conversation that outgrows the model's context window.

  - **Capabilities**
  - [Working with Files](/agent/files): How Jan Agent reads, searches, edits, and runs commands in your project.
  - [Web Search and Fetch](/agent/web): Let the agent look things up when the answer isn't in your project.
  - [Subagents](/agent/subagents): Delegate parts of a task to isolated agents that run in parallel, each with its own context and tool allowlist.
  - [Memory](/agent/memory): Durable facts the agent keeps about your project, stored as files you can read and edit.
  - [Skills](/agent/skills): Reusable procedures the agent loads when a task calls for them.
  - [Skill Invocation](/agent/skill-invocation): The two ways a skill runs - typed by you, or fired by the agent - and how to choose.
  - [Plugins](/agent/plugins): Install skills from git repos or a marketplace as plugins.

  - **Configuration**
  - [Providers](/agent/providers): Connect Jan Agent to a model - cloud, Tokamak, or your own OpenAI-compatible endpoint.
  - [Contributing a Provider](/agent/contributing-providers): How to add a new model provider to Jan Agent - from a plain OpenAI-compatible endpoint to a custom wire API.
  - [Project Config](/agent/project-config): AGENTS.md and agent.toml - per-project instructions, model, budget, and tool policy.
  - [Tool Permissions](/agent/permissions): The approval model, capabilities, and how to widen or narrow what the agent can reach.
  - [Hooks](/agent/hooks): Shell commands the agent runs at fixed points of a run, to enforce a policy the model cannot ignore.
  - [MCP](/agent/mcp): Use your Model Context Protocol servers from the Jan Agent console.
  - [Hide Secrets](/agent/hide-secrets): Keep API keys, tokens and passwords out of the requests Jan Agent sends to a model provider, while tools still run with the real values.
  - [Telemetry (OpenTelemetry)](/agent/telemetry): Export Jan Agent usage metrics and events over OTLP to your own collector, so a team can see tokens, cost, tool activity and errors in one place.

  - **Embedding**
  - [Jan ADK](/agent/adk): Add an AI agent to your app. Start with a prompt, then give the agent functions your code can run.
  - [Nightly ADK Installation](/agent/adk-nightly): Install prebuilt JavaScript and Python ADK previews without npm or PyPI publication, a repository clone, or a Rust build.
  - [JavaScript Quickstart](/agent/adk-ts): Install the preview ADK, connect a model, and stream your first Jan Agent response from Node.js.
  - [Python Quickstart](/agent/adk-python): Install the preview ADK, connect a model, and stream your first Jan Agent response from Python.
  - [Host Tools](/agent/adk-tools): Give the agent a function in your app. Declare what it does, run the callback, and return a result.
  - [Process & Failures](/agent/adk-runtime): Keep control of an embedded Jan runtime: close it, interrupt work, answer permissions, and handle failures without repeating side effects.
  - [Protocol Records](/agent/protocol): The wire reference for custom Jan clients. Choose RPC or stream-json, then use the generated schemas for record shapes.
  - [Editors (ACP)](/agent/acp): Drive Jan Agent from Zed, JetBrains and other editors over the Agent Client Protocol with the experimental `jan acp` server.

  - **Reference**
  - [CLI Reference](/agent/cli): Every Jan Agent flag and subcommand, including headless runs for scripts and CI.