FluxyChat Docs

FluxyChat stores connector metadata in KV and indexed documents in D1 (support_knowledge_base). Start with a URL or manual paste connector; Notion, Confluence, Google Drive, Intercom, and Zendesk appear in the dashboard as source types — sync via pasted content or your own OAuth worker until marketplace OAuth ships.

Dashboard

Open Knowledge base in the console:

Add source — pick type and name (URL sources need a fetchable https link).
Sync — URL sources fetch and ingest; other types use pasted Markdown/text.
Search — keyword search over published articles.
Attach RAG to agents with POST /admin/kb/rag or the KB + RAG cookbook.

Admin API

All routes require an admin JWT.

Method	Path	Purpose
GET	`/admin/kb/sources`	List connectors
POST	`/admin/kb/sources`	Create connector `{ type, name, config? }`
DELETE	`/admin/kb/sources/:id`	Remove connector metadata
POST	`/admin/kb/sources/:id/sync`	URL fetch or `{ content, title? }` ingest
POST	`/admin/kb/ingest`	Direct ingest `{ sourceId, title, content }`
GET	`/admin/kb/search?q=`	Search index
POST	`/admin/kb/rag`	`{ query, maxResults? }` → `synthesizedPrompt`

Example: URL connector

curl -X POST "$WORKER/admin/kb/sources" \
  -H "Authorization: Bearer $JWT" \
  -H "Content-Type: application/json" \
  -d '{"type":"url","name":"Runbooks","config":{"url":"https://docs.example.com/runbooks"}}'

curl -X POST "$WORKER/admin/kb/sources/kbs_abc123/sync" \
  -H "Authorization: Bearer $JWT"

RAG for agent invoke

curl -X POST "$WORKER/admin/kb/rag" \
  -H "Authorization: Bearer $JWT" \
  -H "Content-Type: application/json" \
  -d '{"query":"How do I reset MFA?"}'

Use synthesizedPrompt as extra system context before POST /agents/:id/invoke.

SDK (client-side dev)

For local prototyping without the Worker:

import { createKnowledgeBase } from "@fluxy-chat/sdk";

const kb = createKnowledgeBase();
const source = kb.addSource({
  type: "notion",
  name: "Wiki",
  connectionConfig: { integrationToken: process.env.NOTION_TOKEN! },
  enabled: true,
  syncIntervalMs: 3_600_000,
});

Production indexing should go through Worker /admin/kb/* so secrets stay server-side.

Knowledge base connectors