Skip to main content
Visual embeds are quote images that get attached to AI responses based on semantic matching. Each embed belongs to a category and stores a GCS image URL plus optional quote text.

Quick facts

  • Base path: /api/v1/visual-embeds
  • MongoDB collections: visual_embeds, visual_categories
  • Auto-seeds from hardcoded manifest on first startup
  • Service caches DB results for 5 minutes

Surface area

See the endpoint pages for request bodies, parameter constraints, and example responses pulled directly from the OpenAPI schema.

Categories

How it works

  1. Chat service generates AI response
  2. Visual Planner (LLM) decides if visual should attach
  3. Visual Embed Service queries visual_embeds by category
  4. Business rules checked in Redis (max 3/session, no repeats)
  5. Visual URL attached to response
Images stored in GCS: gs://handa-visuals/{Category}/{filename.png}