fix: increase WS client thread stack to 16KB
MCP HTTP handler adds body[4096] on stack inside client_thread, pushing total stack usage to 100% of the 12KB allocation. This causes Data Abort (DFAR=0x99) when curl sends MCP initialize.
Increase AGENT_WS_CLIENT_STACK from 12KB to 16KB. RAM impact: +4KB.
Signed-off-by: zhouwenjie1 zhouwenjie1@xiaomi.com
版权所有:中国计算机学会技术支持:开源发展技术委员会
京ICP备13000930号-9
京公网安备 11010802047560号
AI Agent
运行在 openvela 嵌入式系统上的 AI Agent 框架。
多 LLM 后端 · ReAct 工具调用 · 多渠道接入 · 多节点协作 · 语音交互 · 低内存优化
所有运行 openvela OS 的智能设备(手表、眼镜、音箱、耳机等)都可以部署此 AI Agent,设备间通过 OpenClaw Node 协议互联互通。
功能亮点
快速开始
编译
前提条件:下载 openvela 源码
基本使用
支持的 LLM 后端
kimiqwendeepseekglmmimoopenaiclaudeopenrouterLLM Router(多后端路由)
支持最多 4 个后端同时配置,按任务复杂度自动选择最优后端:
内置工具(35+)
web_searchnews_searchfetch_urlread_filewrite_fileedit_filelist_dircron_addcron_listcron_removeanalyze_imagecamera_capturerun_shellget_batteryget_wear_stateget_screen_stateget_heartrateget_stepsvibratefeishu_doc_createfeishu_doc_writefeishu_doc_readfeishu_doc_listfeishu_chat_membersfeishu_send_mentionmusic_searchmusic_playmusic_pausemusic_resumemusic_stopmusic_seekmusic_set_volumemusic_statuslaunch_quickappexit_quickappget_weatherget_current_timemcp_add接入接入通道
ask <text>直接对话详见 接入通道文档。
Skills 系统
10 个内置 Skill,Markdown 格式,对话即可创建新技能:
weatherdaily-briefingskill-creatorremindernote-takertranslatenews-digesttask-managersystem-healthfeishu-test配置
Kconfig 模块开关
所有可选模块均可通过 Kconfig 独立开关,按需裁剪 RAM/Flash 占用:
AI_AGENT_FEISHUAI_AGENT_WEIXINAI_AGENT_MQTTAI_AGENT_NODEAI_AGENT_MCPAI_AGENT_LVGL_UIAI_AGENT_CAMERAAI_AGENT_AUDIO_PREPROCESSAI_AGENT_SKILL_SYNCAI_AGENT_NET_RPMSGAI_AGENT_TEST最小配置(CLI-only):关闭 FEISHU + WEIXIN + MQTT + NODE + MCP + LVGL_UI,预估节省 ~260KB RAM。
配置文件
配置文件位于
/data/agent/config/:SOUL.mdUSER.mdconfig.json文档
Acknowledgments
AI Agent is a derivative work based on MimiClaw, originally created by Ziboyan Wang and contributors under the MIT License.
AI Agent reimplements and extends MimiClaw’s AI agent architecture for the Vela/NuttX platform, adding multi-channel support (Feishu/WeChat/MQTT/Voice), multi-device collaboration (OpenClaw Node), MCP protocol integration, and optimizations for resource-constrained wearable devices.
We gratefully acknowledge the MimiClaw project and its contributors for the foundational work that made AI Agent possible.
License
Apache-2.0
This project includes code derived from MimiClaw (MIT License, Copyright (c) 2026 Ziboyan Wang). See NOTICE for details.