Observer
Observer is built around a simple "when X then Y" pattern. Each micro-agent has a system prompt, a loop interval and sensors such as $SCREEN, $SCREEN_OCR, $CAMERA, $CLIPBOARD or $MICROPHONE (transcribed with Whisper via transformers.js). A small model looks at the input every cycle and a short JavaScript snippet decides what to do with its answer.
The tool layer is where it stands out: sendEmail, sendDiscord, sendTelegram, sendPushover, SMS, WhatsApp and even automated phone calls, plus clip recording, per-agent text and image memory, and starting or stopping other agents. Some channels (calls, SMS, WhatsApp) require whitelisting by the project, and SMS is blocked for US/Canada numbers.
Models run locally through Ollama, llama.cpp, vLLM or any OpenAI-compatible endpoint; the desktop app is described as the most stable option, and there is also a hosted web app. The Docker setup is marked deprecated. License is AGPL-3.0.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-20
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 17
- Latest release
- v3.1.1 (2026-10-16)
- Open issues
- 0
- Recent issues with at least one comment
- 20/20
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Compatibility
Use cases
Alternatives
Comments
No comments yet. Be the first!