Docs

Odyssey is a project-scoped memory system that sits between your data and your model. It handles ingestion, retrieval, structure, and debugging, all in one place.

Instead of treating retrieval as a black box, Odyssey makes it visible, editable, and testable.

Most AI systems rely on retrieval that cannot be inspected, controlled, or debugged. When something goes wrong, you cannot see what the model was given or why it was selected.

Odyssey solves this by exposing the full retrieval layer:

• see exactly what memory was retrieved
• inspect source documents and chunks
• modify structure to influence future results
• debug behavior without calling a model

Odyssey follows a continuous loop:

1. Ingest memory
2. Structure memory (entities, relations, tags)
3. Retrieve context
4. Inspect what was returned
5. Refine memory to improve future retrieval

Every surface in Odyssey maps to one part of this loop.

Odyssey allows you to test retrieval without calling a model.

This means you can:

• see exactly what context would be sent to a model
• debug retrieval without burning tokens
• export retrieved documents for inspection

This is the fastest way to understand and fix retrieval behavior.

Memory in Odyssey is organized as:

• Documents → raw inputs
• Chunks → indexed units used for retrieval
• Entities → key concepts
• Relations → connections between entities
• Tags → additional control signals

Entities and relations are both automatically generated and manually editable.

Odyssey uses a graph layer to improve retrieval.

It considers:

• entities
• relations
• neighbor connections
• tags and recency

Users can enable or disable graph usage and influence it by editing structure.

• tags
• entities
• relations
• memory content
• graph usage (on/off)

Instead of tuning ranking parameters directly, you influence retrieval by shaping the memory and structure itself.

Add text into project-scoped memory and receive a successful ingest response when Odyssey indexes it.

{
  "text": "Odyssey supports project-scoped retrieval and graph-aware ranking for AI systems.",
  "source": "api_v1",
  "source_ref": "doc_hello_world",
  "tags": ["retrieval", "graph", "docs"],
  "doc_title": "Odyssey Retrieval Notes"
}

{
  "status": "ok",
  "mode": "text",
  "chunk_id": "chk_7f3a91c2"
}

Query project memory and return ranked context in Odyssey’s retrieval format.

{
  "query": "how does Odyssey use graph-aware retrieval?",
  "top_k": 2,
  "filters": null
}

{
  "items": [
    {
      "id": "mem_8f3a",
      "text": "Odyssey enables project-scoped retrieval with optional graph-aware ranking.",
      "summary": "Project-scoped retrieval with graph support.",
      "snippet": "project-scoped retrieval with graph support",
      "tags": ["retrieval", "graph"],
      "source": "file",
      "doc_title": "retrieval-overview.pdf",
      "page_no": 2
    },
    {
      "id": "mem_c29b",
      "text": "Use the retrieve endpoint to query indexed memory and return ranked context.",
      "summary": "Query indexed memory through the retrieve API.",
      "snippet": "query indexed memory through the retrieve API",
      "tags": ["api", "memory"],
      "source": "text",
      "doc_title": "api-notes"
    }
  ],
  "count": 2,
  "meta": {
    "project_id": "YOUR_PROJECT_ID",
    "graph_used": true
  }
}

Use API keys to access ingest and retrieval endpoints.

x-api-key: YOUR_ODYSSEY_API_KEY

Model keys are used only in the Interface to connect your own model.

• AI assistants with persistent memory
• internal knowledge systems
• agent memory infrastructure
• retrieval debugging and evaluation