Files
stack/docs/scratchpads/ms22-conversation-archive.md
Jason Woltje de37e7be90
All checks were successful
ci/woodpecker/push/orchestrator Pipeline was successful
ci/woodpecker/push/web Pipeline was successful
ci/woodpecker/push/api Pipeline was successful
feat(api): add conversation archive with vector search (MS22-DB-004, MS22-API-004)
- Add ConversationArchive Prisma model with pgvector(1536) embedding field
- Migration: 20260228000000_ms22_conversation_archive
- NestJS module at apps/api/src/conversation-archive/ with service, controller, DTOs, spec
- POST /api/conversations/ingest — ingest session logs, auto-embed via EmbeddingService
- POST /api/conversations/search — vector similarity search with agentId filter
- GET  /api/conversations — paginated list with agentId + date range filters
- GET  /api/conversations/:id — fetch full conversation including messages
- Register ConversationArchiveModule in app.module.ts
- 8 unit tests, all passing (vitest)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-02-28 20:11:39 -06:00

1.6 KiB

MS22 — Conversation Archive Module

Objective

Implement ConversationArchive module: ingest OpenClaw session logs, store with vector embeddings for semantic search.

Deliverables

  1. ConversationArchive Prisma model
  2. NestJS module at apps/api/src/conversation-archive/
  3. Endpoints: ingest, search, list, get-by-id
  4. Register in app.module.ts
  5. Migrate, lint, build, commit

Plan

  • Add model to schema.prisma (end of file)
  • Add relation to Workspace model
  • Create module structure: dto/, service, controller, spec, module
  • Use EmbeddingService from knowledge module (import KnowledgeModule or just PrismaModule + embed inline)
  • Follow pattern: AuthGuard + WorkspaceGuard + PermissionGuard
  • Endpoint prefix: conversations (maps to /api/conversations)
  • Vector search: $queryRaw with <=> operator (cosine distance)

Assumptions

  • ASSUMPTION: Embedding is stored inline on ConversationArchive (not a separate table) — simpler and sufficient for this use case, matches MemoryEmbedding pattern
  • ASSUMPTION: Import KnowledgeModule to reuse EmbeddingService (it exports it)
  • ASSUMPTION: messageCount computed server-side from messages array length on ingest
  • ASSUMPTION: Permission level WORKSPACE_MEMBER for ingest/search, WORKSPACE_ANY for list/get

Progress

  • Schema model
  • Migration
  • DTOs
  • Service
  • Controller
  • Spec
  • Module
  • app.module.ts registration
  • Lint + build
  • Commit

Risks

  • EmbeddingService exports from knowledge.module — need to import KnowledgeModule
  • Migration requires live DB (may need --skip-generate flag if no DB access)