JiuwenClaw is a self-hosted personal AI agent written in Python that connects large language models to the communication apps you already use. You run it on your own infrastructure so data stays under your control, and you talk to it through a web front end, through Feishu once a channel is configured, or, on Huawei phones, through the Xiaoyi assistant via native Xiaoyi Open Platform integration. Python 3.11 to 3.13 is supported.
The agent is designed around task handling and self-improvement. It takes new tasks, interruptions mid-flow, and changing requirements and schedules the work in order; when you express dissatisfaction or an error occurs, it refines the relevant skills from your feedback. Heartbeat tasks let it wake on a schedule to work through your to-do items unattended.
Model access covers Huawei Cloud MaaS and other mainstream model platforms. Documentation covers configuration and workspace layout, a skill system for custom skills, channel setup, CLI commands, scheduled tasks, memory, browser automation, task planning, skill self-evolution, and context compression and unloading.
Features
- Multi-platform access: web chat at a local port, Feishu, and the Xiaoyi assistant on Huawei phones
- Model platforms: Huawei Cloud MaaS plus other mainstream providers
- Self-hosted deployment: runs on your own machine or server with a jiuwenclaw-init setup step and a jiuwenclaw-start launcher
- Task scheduling: handles added tasks, interruptions, and shifting requirements in order
- Scheduled heartbeat tasks: wakes on a timer to execute queued to-do items
- Skill self-evolution: rewrites skills in response to user feedback or errors
- Memory and browser automation: persistent memory and web-browsing capabilities
- Context compression: compresses and unloads context for long sessions
