Episodic Archiver

This skill analyzes full conversation transcripts, embeds them for search, and categorizes turns.

Usage

# Save a JSON transcript to episodic memory
.agents/skills/episodic-archiver/run.sh archive <transcript.json>

Input Format

The JSON file should be a list of messages or an object with a messages list:

{
  "session_id": "task_123",
  "messages": [
    {
      "from": "Coordinator",
      "to": "Worker",
      "message": "Please fix the bug...",
      "timestamp": 1234567890
    },
    ...
  ]
}

Storage

Data is stored in ArangoDB collection agent_conversations with:

• embedding: Vector representation (768d)
• category: LLM-derived tag (Task, Question, Solution, etc.)

Prerequisites

• ArangoDB credentials in .env (e.g., ARANGO_URL, ARANGO_DB, ARANGO_USER, ARANGO_PASS).
• Embedding service running (recommended): EMBEDDING_SERVICE_URL=http://127.0.0.1:8602
• Falls back to local graph_memory embedding model if service unavailable.
• Optional: CHUTES_API_KEY or OPENAI_API_KEY (for categorization).