Journey RAG

Journey: RAG

Stage 3g of the guided journey. Create RAG collections and ingest initial documents.

When to use

• Dispatched by journey when current_stage: rag.
• Directly via /butterbase-skills:journey-rag.
• Skipped (annotated (n/a)) if the plan has no RAG section.

Preflight

If docs/butterbase/03-preflight.md is missing, older than 24 hours, or 00-state.md has app_id: null, invoke butterbase-skills:journey-preflight first. Wait for it to return successfully before proceeding.

Inputs

• docs/butterbase/02-plan.md — the RAG section.
• docs/butterbase/00-state.md — for app_id.

Procedure

• Refresh docs. Call butterbase_docs with topic: "rag". For ingestion + retrieval patterns, also WebFetch https://docs.butterbase.ai/ai/rag. Skip if cache is fresh.

• Read the RAG section. Print it back: "About to set up RAG: collections=<list>. Proceed?". Wait for yes.
• Invoke butterbase-skills:rag-dev via the Skill tool with the RAG plan and app_id. The wrapped skill calls manage_rag_content action: create_collection, then ingest_document for any seed sources the user provides.
• Smoke: call rag_query with a representative question and show the user the top hit.
• Append one line to docs/butterbase/04-build-log.md:

<ISO timestamp> rag manage_rag_content ok

• Tick - [x] rag in 00-state.md, set current_stage: to the next unchecked stage.
• Return to journey orchestrator (or ask "Continue to the next stage? (yes/no)").

Outputs

• Created RAG collections, ingested seed docs.
• One line in 04-build-log.md.

Anti-patterns

• ❌ Ingesting huge document sets in a single call — chunk and poll get_document_status.
• ❌ Skipping the smoke rag_query — silent embedding failures are common.