Skip to content

selfhost/rag: boot-time preflight for an embed-capable provider + compose-file RAG hints #8765

Description

@JSONbored

Context

Parent: #8757. RAG audit finding: with LOOPOVER_REVIEW_RAG=true and a provider chain of only CLI-subscription providers (e.g. AI_PROVIDER=claude-code) and no AI_EMBED_BASE_URL, every embed call throws (claude_code_no_embed, src/selfhost/ai.ts:965) and the index never populates — surfaced only as per-batch runtime ERROR logs (rag.ts:437-449). The review-CLI path already gets a fail-LOUD boot preflight (#1566 pattern in src/server.ts); RAG's embed dependency does not.

Requirements

  • At boot, when RAG is enabled and neither AI_EMBED nor any embed-capable member of the AI chain exists, log the same class of loud structural warning as selfhost_ai_cli_missing (do not crash — degrade exactly as today, but say so once, at startup).
  • Add AI_EMBED_BASE_URL/AI_EMBED_MODEL mentions to docker-compose.yml's qdrant/ollama profile hint comments (today only .env.example line ~791 explains it).

Deliverables

  • The preflight + test (embed-capable present: silent; absent: warns once).
  • The compose hint comments.

Test Coverage Requirements

99%+ patch coverage, branch-counted on the preflight.

Metadata

Metadata

Assignees

Labels

maintainer-onlyOwner-only work — yields no Gittensor points.

Projects

No projects

Relationships

None yet

Development

No branches or pull requests

Issue actions