Retrieval Sources (RAG)

What it does

Retrieval Sources (RAG) registers the systems Arbitex can retrieve grounding content from — a vector database, web crawler, S3 bucket, Confluence space, or SharePoint site — through a typed connection form rather than raw JSON.

Registration is live; automatic grounding is planned

Sources you register here are versioned and synced to the deployment today — this part is fully live. Automatically using a registered source to ground requests at inference time is planned and not yet wired; registering a source does not, by itself, change how any request is answered yet.

Configure in the UI

Models & Routing → Retrieval Sources (RAG)Citadel synthetic data on a non-production Arbitex demo instance. Never client data.

1. Open Models & Routing → Retrieval Sources (RAG).

2. Set a Source ID — a stable identifier you cannot change once created — a Name, and a Source type: Vector database, Web crawler, S3 bucket, Confluence, or SharePoint.

3. Enter the Endpoint the source is reachable at, and optionally an Index / collection name.

4. Tune retrieval with Top K (how many results to retrieve) and Minimum score (the relevance threshold), then toggle Enabled and save.

5. The table lists every source’s Type, Connection (endpoint and index), Grounding settings (top_k / min_score), and Status — with Edit and Delete row actions.

No credential field in this form

The connection form has no dedicated authentication/credential field — only an Endpoint. If your source requires authentication, that is not yet represented on this page.

API equivalent

curl -X POST "https://admin.arbitex.ai/api/v1/admin/models-routing/retrieval-sources/list" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"du_id": null}'

curl -X POST "https://admin.arbitex.ai/api/v1/admin/models-routing/retrieval-sources/set" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"du_id": null, "objects": [{"object_key": "knowledge-base-primary", "name": "Product knowledge base", "source_type": "vector_db", "endpoint": "https://vector-db.internal:6333", "top_k": 5, "min_score": 0.75, "enabled": true}]}'

set is replace-all for the segment — every add, edit, or delete re-authors the complete source set as one revision. There is no test-connection action on this page today.

Plan & availability

Retrieval Sources (RAG) is available on every plan.

Deployment notes

The Endpoint must be reachable from wherever the request is served. On a Hybrid Outpost, a vector database or S3-compatible endpoint inside your network works well; Web crawler, Confluence, or SharePoint sources that need internet reachability will not function from an air-gapped outpost without an internal mirror.

Related

• Model Catalog — models that will eventually consume grounded context from these sources.
• MCP Servers — another typed connection surface, for tool servers rather than retrieval sources.