openworker vs picobot
Andrew Ng's local-first desktop AI coworker: give it an outcome and it works across your files, terminal and 25+ apps — Slack, Jira, Notion, Gmail — then hands back the finished deliverable. — versus — A self-hosted personal AI agent in a single ~9MB Go binary — persistent memory, 16 tools + MCP, skills, cron/heartbeat, and Telegram/Discord/Slack/WhatsApp channels. Runs on a $5 VPS.
Same 'agent that just runs your errands' idea, different host: picobot is a 9MB Go binary on a $5 VPS reachable from Telegram or Slack, openworker a local-first desktop coworker driving your files, terminal and 25+ apps.
| openworker | picobot | |
|---|---|---|
| Stars | 15k | 1.3k |
| Forks | 2.1k | 166 |
| Language | Python | Go |
| License | MIT | MIT |
| Last activity | today | 4 months ago |
| Topics | agents | agents |
| Curated connections | 5 | 3 |
openworker — the curator's take
Use it when you want an agent that produces artifacts rather than chat: it plans, works across your desktop and connected apps, and stops for approval before sending, writing or running anything. The agent loop, conversations, connector tokens and model keys all stay in the app's local store; bring your own key for 13-plus providers or point it at Ollama. Scheduled automations park their questions in an inbox instead of acting alone. Caveats: it's an open beta desktop app (macOS signed and notarized, Windows builds not yet code-signed), a small cloud service brokers connector OAuth unless you paste credentials by hand, and 'finished work' is only as good as the model you plug in.
picobot — the curator's take
The anti-bloat statement piece: zero dependencies, ~10MB RAM idle, instant cold start — an always-on agent with ranked memory recall, background subagents, a natural-language HEARTBEAT.md cron, and self-authored skills ('create a skill for checking weather' → it writes the markdown), reachable from your phone via four chat channels. Runs on a Raspberry Pi or Termux on an old Android. Any OpenAI-compatible endpoint, including Ollama. Its own README names the target: OpenClaw's power without the 500MB container. NOT a coding harness and single-user by design — this is your personal daemon, not a team platform; complex multi-agent workflows will outgrow it fast, which is rather the point.