CraftBot

What is CraftBot?

CraftBot is a personal AI assistant that runs on your own machine and is designed to interpret tasks, plan actions, and execute them to help achieve goals. It is intended to work continuously so it can proactively help you plan and initiate tasks based on your preferences and objectives.

The project supports multiple language-model providers and includes integrations, a memory system, and an extensible framework for extending what the agent can do through MCP (Model Context Protocol) and custom skills.

Key Features

• Bring Your Own Key (BYOK) LLM provider system: Select and switch between supported LLM providers including OpenAI, Google Gemini, Anthropic Claude, BytePlus, and local Ollama.
• Memory system with daily consolidation: Distills and consolidates events from the day at midnight to support ongoing personalization.
• Proactive agent behavior (with approval): Learns preferences, habits, and life goals; can plan and initiate tasks, with approval noted as part of the workflow.
• External tools integration with OAuth support: Connect to services including Google Workspace, Slack, Notion, Zoom, LinkedIn, Discord, and Telegram (with more to come noted).
• MCP integration for extended capabilities: Uses Model Context Protocol to extend agent capabilities with external tools and services.
• Skills framework: Extensible skill framework with built-in skills for planning tasks, research, code review, and git operations.

How to Use CraftBot

1. Install prerequisites: Use Python 3.10+ and clone the repository. You’ll also need an API key for your chosen LLM provider (OpenAI, Gemini, or Anthropic are mentioned).
2. Install dependencies: Run python install.py from the cloned CraftBot directory.
3. Start the agent: Run python run.py. The first run guides you through setting up your API keys.
4. Choose an interface mode:
   • Browser (default): python run.py (recommended; requires Node.js 18+).
   • TUI: python run.py --tui (no Node.js dependency mentioned).
   • CLI: python run.py --cli.
   • GUI (experimental): python run.py --gui install.py --gui (GUI mode is explicitly described as experimental; issues may occur during switching).
5. Start interacting: Use natural language to talk to the agent and view available commands with /help. You can also connect integrations like Google, Slack, and Notion after setup.

Use Cases

• Personal task planning and initiation: Ask the agent to plan multi-step tasks and initiate work aligned with your preferences and life goals, with the workflow described as including approval.
• Cross-tool assistance across work apps: Use the integrated connections (e.g., Google Workspace, Slack, Notion, Zoom, LinkedIn, Discord, Telegram) to help with tasks that span multiple services.
• Extending capabilities via MCP: Add or connect external tools and services through MCP so the agent can interact with additional capabilities beyond its built-in set.
• Developer-oriented workflow help: Use built-in skills such as research, code review, and git operations to support software work directly from the assistant.
• Choosing a suitable interface for your environment: Run in browser mode for a web interface, use TUI/CLI for lightweight terminal workflows, or attempt the experimental GUI mode if you want visual feedback.

FAQ

• What models can CraftBot use? CraftBot supports a BYOK LLM system with multiple providers, including OpenAI, Google Gemini, Anthropic Claude, BytePlus, and local Ollama.

• Do I need Node.js? Node.js 18+ is noted as optional and only required for the browser interface; if Node.js is not installed, the installer provides instructions or you can use TUI mode.

• Is the GUI mode stable? No—GUI mode is described as experimental, and you may encounter issues when the agent switches to GUI mode. The project notes active improvement.

• Can I extend what CraftBot can do? Yes. The repository lists support for MCP integrations and a skills framework, both intended to extend agent capabilities.

• Which integrations are supported out of the box? The page lists integrations for Google Workspace, Slack, Notion, Zoom, LinkedIn, Discord, and Telegram, with embedded credentials and OAuth support.

Alternatives

• Local, tool-using AI agents built with frameworks: If you want a similar “agent runs on your machine” setup but with a different ecosystem, look for frameworks that support tool calling, memory, and provider switching.
• Chat-based assistants with external tool connections: If your goal is simpler chat with integrations (rather than autonomous planning/execution), consider assistants that focus on conversational workflows with connected apps.
• Command-line productivity automation: For users who prefer automation without an AI planning loop, alternatives include task runners and automation tools that orchestrate actions based on predefined workflows.
• MCP-capable agent setups: If your main need is extending capabilities via MCP, compare other MCP-enabled agent architectures or implementations to match your preferred UI and tool set.