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., ARANGOURL, ARANGODB, ARANGOUSER, ARANGOPASS).
• Local embedding model availability (graph_memory’s embedding stack).
• Optional but recommended: CHUTESAPIKEY or OPENAIAPIKEY (for categorization).