Agent bridge
How to connect RAG maintenance to an agent or webhook
A simple model for sending RAG repair events to an agent, webhook, email inbox, or future chat adapter.
The easiest way to connect RAG maintenance to an agent is to send signed webhook events for care-cycle results and expose signed inbound actions for scans, status checks, and repair approval.
What events should a RAG care agent receive?
The first useful events are care-cycle completed, care-cycle failed, source stale, source broken, retrieval test failed, repair created, and repair approved.
What actions should an external agent be allowed to take?
The safe MVP action set is run scan, get workspace health, list repairs, approve repair, ignore repair, and reopen repair. Production writes should wait for repair previews and approval controls.
- Use signed requests so the receiver can trust ChunkFunk events.
- Use workspace-scoped tokens or channel signatures for inbound actions.
- Log delivery history so users can retry failed events.
- Keep email as the fallback when no external channel is connected.
Why start with a generic webhook instead of one chat app?
A generic webhook supports OpenClaw, Hermes-style agents, custom scripts, Make, Zapier, n8n, and future Telegram, Discord, Slack, or WhatsApp adapters without forcing one communication surface.
Common questions
Should ChunkFunk be Telegram-first?
No. The better MVP is channel-first: webhook and agent bridge first, email fallback, then chat adapters after the core repair loop proves useful.
Can an agent approve RAG repairs?
Yes, if it uses a scoped API token or signed channel request. Silent production writes should still wait for explicit approval and preview safeguards.
Why use signed webhook requests?
Signing lets external agents verify that a repair or care-cycle event came from ChunkFunk and was not altered in transit.