Give your AI a memory

The write path entry point. Accepts conversations, summaries, queries, sentiment, and tool calls. Returns immediately — processing happens asynchronously.

# Request — Bearer or X-App-ID auth
{
  "user_id": "user-123",
  "interaction_type": "chat_log",
  "content": "User discussed project timeline...",
  "response": "AI suggested Q2 deadline...",
  "metadata": {
    "meeting_id": "mtg-456",
    "project": "Widget 2.0"
  }
}

# Response 200
{
  "status": "queued",
  "message": "Memory update received for async processing"
}

The hot path. Matches entities in the user's text against their knowledge graph and returns a pre-formatted context block. Target: <10ms. No LLM call involved.

# Request
{
  "user_id": "user-123",
  "text": "What's the status of Widget 2.0?",
  "max_hops": 2
}

# Response 200
{
  "context": "Sarah is a PM at Acme Corp.
  Prefers direct answers. Working on Widget 2.0.
  Decided on Nova 3 for transcription.
  Samples due Thursday. David (CTO) is sponsor.",
  "matched_entities": ["Widget 2.0", "Sarah"],
  "patch_count": 7
}

Returns all facts, action items, and patch connections for a user. Supports filtering by category and incremental sync via since timestamp.

# Response 200
{
  "user_id": "user-123",
  "facts": [
    {
      "patch_id": "uuid",
      "fact": "Product Manager at Acme",
      "category": "identity",
      "patch_type": "identity",
      "source": "inferred",
      "connections": [
        {
          "to_patch_id": "uuid",
          "role": "works_on",
          "label": "Widget 2.0"
        }
      ]
    }
  ],
  "action_items": [...],
  "server_time": "2025-01-15T..."
}

Generates a force-directed graph visualization of the user's quilt. Returns SVG, PNG, or interactive HTML. Color-coded by patch type with edge coloring by relationship role.

Let users correct extracted facts. Accepts fact and category fields. Marks the patch as user-declared.

Permanently remove a single patch from the user's quilt.

Complete data deletion. Removes all patches, entities, relationships, and cached data for a user.

Rename a speaker or entity across all patches, relationships, and the entity index. Useful when the system labels someone as "Speaker 4" and the user knows their real name.

Returns all projects for a user with status and patch counts.

Create a named project. Patches and entities can be scoped to projects for organized context retrieval.

Rename a project or change its status to archived. Archiving cascades to child patches.

Pass a prompt template with [[placeholder]] syntax. CQ substitutes values from the user's profile. Supports defaults via [[key|fallback]].

# Request
{
  "user_id": "user-123",
  "template": "You are helping [[name|a user]],
  a [[role]]. They prefer [[communication_style]].
  Active project: [[current_project|none]]."
}

# Response 200
{
  "enriched_prompt": "You are helping Sarah,
  a Product Manager. They prefer concise,
  direct answers. Active project: Widget 2.0.",
  "used_variables": ["name", "role", "communication_style", "current_project"],
  "missing_variables": []
}

Returns the cached user profile from Redis. Supports ?keys=key1,key2 to filter specific fields. Useful for building personalized UIs.

Hydrate a user's profile and entity index from PostgreSQL into Redis. Call at session start to ensure the first /v1/recall is a cache hit. Completes in <50ms.

Returns service status and version. No authentication required.