// the find
HKUDS/nanobot
Lightweight, open-source AI agent for your tools, chats, and workflows.
nanobot is a Python-based personal AI agent that wraps LLM providers (OpenAI, Anthropic, local models) with persistent memory, tool execution, scheduled tasks, and bridges to chat platforms like Telegram, Slack, Discord, and WeChat. It targets developers who want a self-hosted, programmable agent they can extend without adopting a heavyweight framework. At 44k stars it's clearly struck a nerve, though the project is only about 4 months old.
The channel coverage is genuinely impressive — Telegram, Slack, Discord, Teams, WeChat, Feishu, Matrix, Signal, email, and WhatsApp in a single codebase without pulling in a separate integration platform. The provider abstraction is clean enough that adding a new LLM backend reportedly takes two steps, and the fallback model chain (`fallbackModels`) is a practical feature most agent frameworks skip. The WebUI ships bundled inside the wheel with no separate build step, which removes a real deployment headache. The test suite is substantial and well-organized with separate directories for agent, channels, and bus concerns.
The daily changelog cadence (shipping every single day for months) is a red flag for stability — this is a project that moves fast and breaks things, and the session/memory system has been redesigned at least twice. The 'lightweight' framing in the README doesn't match what's actually in the repo: 15+ chat channel implementations, a full WebUI, image generation, MCP, subagents, and cron scheduling is not a small core. The memory system's durability under failure (session atomicity, mid-run crashes) is an open concern given the frequent fixes mentioned in changelogs. Configuration is JSON-heavy and has accumulated multiple overlapping patterns (direct provider/model vs. named presets), suggesting the API surface hasn't settled yet.