Autonomous research agent for Claude Code. Reads research requests from Slack, tracks findings in kendb, expands research when idle, and self-improves.
- Slack authentication via clacks: run
clacks auth loginbefore first use - clacks and ken are installed automatically on first run
/plugin marketplace add zomglings/ongo
/plugin install ongo@ongo
/ongo
Options:
--channel <id>— Slack channel to listen on (default: self-DM)--interval <seconds>— poll interval in seconds (default: 3)--idle— only respond to messages; do not expand research autonomously
Ongo runs as a game loop inside your Claude Code session:
- Poll — checks Slack for new messages every few seconds
- Process — interprets messages as natural language (research requests, strategy updates, maintenance triggers)
- Expand — when idle, picks a random topic from kendb and researches it further
- Self-improve — every 24 hours, runs kendb maintenance, checks for dependency updates, and reflects on its own behavior
All research is tracked in kendb with publications, relationships, and notes.
Tell ongo how to focus its research via Slack:
- "Focus more on cryptography"
- "Prefer papers published after 2020"
- "Ignore machine learning"
These are stored as ongo-exploration entries in kendb and consulted during idle expansion.
Every 24 hours (or on request), ongo:
- Maintains kendb — deduplicates, fills relationship gaps, produces surveys, evolves kinds
- Updates dependencies — checks for new ken and clacks releases
- Modifies itself — edits its own skill instructions based on what's working
All self-improvement attempts are tracked in kendb via the ongo-self-improvement kind.
Test locally:
claude --plugin-dir /path/to/ongo
MIT