Engram Engine Capabilities for AI Agents

ATTENTION AI ASSISTANT / IDE AGENT: This document defines the high-level capabilities, scripts, and context-awareness patterns available in this repository. Use these tools to perform complex tasks for the user.

1. Story & Visual Generation (Sage Meridian)

The system includes a dedicated “Storyteller” agent (Sage) capable of generating rich narratives and accompanying visuals (via Imagen 3.0 / Gemini).

  • Capability: Generate a technical story or narrative about a topic, complete with a generated visual asset.
  • Trigger Script: backend/scripts/simulate_sage_story.py

  • Usage:

      python -m backend.scripts.simulate_sage_story "Your Topic Here"

    This script generates the story (Claude/LLM), creates a visual (Gemini/Imagen), saves artifacts to docs/stories, and ingests the content into Zep memory.

2. Tri-Search (Memory & Context)

The “Engram Engine” uses a three-layer search architecture to retrieve context. You can utilize these patterns when building features or debugging.

  • Architecture:
    1. Keyword Search: Precise matching (Zep Messages).
    2. Vector Search: Semantic understanding (Embeddings).
    3. Knowledge Graph: Relationship traversal (Zep/Graphiti).
  • Relevant Code:
    • backend.memory.client: Core client for all memory interactions.
    • backend.etl.ingestion_service.IngestionService: Orchestrates the writing of data to all 3 layers.
    • ingest_text(text, filename, ...): API to ingest raw text into the Tri-Search engine.

3. Data Ingestion Connectors

To bring data into the system, use the specific connectors found in backend/etl/connectors:

  • Wiki: wiki.py (URL -> Engram)
  • Tickets: ticket.py (JSON -> Engram)
  • Code: git_repo.py (Repo -> Engram)

4. Operational Context

  • Environment: Azure (Primary), Local (Docker).
  • Auth: Managed via .env files.
    • Run test_gemini_import.py to debug Gemini/Google env issues.