python-sdk/pkg/hanzo-memory/DOCS.md

Memory API Documentation

Introduction

Memory API provides long-term memory and contextual knowledge capabilities for AI applications, enabling systems to:

• Remember past interactions with users
• Maintain context across sessions
• Retrieve relevant information from previous conversations or a dedicated knowledge base

These capabilities support more personalized, contextually aware, and human-like AI experiences.

Authentication

• Production: All endpoints require an API key provided via the x-api-key HTTP header or set in the HANZO_API_KEY environment variable.
• Local Development: Authentication can be disabled when running locally by setting DISABLE_AUTH=true or omitting the header.

Note: All endpoints are RESTful and expect JSON request bodies with lowercase key names (e.g., userid, messagecontent). The userid parameter is mandatory for most endpoints; it partitions data per user to ensure multi-tenancy and security.

Core Memory API

1. POST /v1/remember

Retrieve & Store Memories

• Description: Retrieves relevant memories for the incoming message, enqueues the message for storage, and optionally filters results via an LLM.
• Authentication: API key required in header Authorization: Bearer <apikey> or JSON field apikey.

Request Parameters

Name	Type	Required	Description
apikey	string	Yes	Your API key.
userid	string	Yes	Unique user identifier.
messagecontent	string	Yes	Message text for retrieval and storage.
additionalcontext	string	No	Extra context to improve retrieval or LLM filtering.
strippii	boolean	No	Anonymize PII during storage (default: false).
filterresults	boolean	No	Use LLM to filter to top 3 memories (default: false).
includememoryid	boolean	No	Return objects with content & memoryId (default: false).

Example Request

POST /v1/remember
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

{
  "userid": "user-123",
  "messagecontent": "I prefer dark mode interfaces",
  "filterresults": true,
  "includememoryid": true
}

Example Response

HTTP/1.1 200 OK
Content-Type: application/json

{
  "userid": "user-123",
  "relevant_memories": [
    {"content": "User previously chose dark theme.", "memory_id": "mem_abc123"}
  ],
  "memory_stored": true,
  "usage_info": {"current": 15, "limit": 1000}
}

---

2. POST /v1/memories/add

Add Explicit Memories

• Description: Directly adds one or more memory strings, bypassing importance analysis.

Request Parameters

Name	Type	Required	Description
apikey	string	Yes	Your API key.
userid	string	Yes	Unique user identifier.
memoriestoadd	string/array	Yes	Single string or array of strings to add.

Example Request

POST /v1/memories/add
{
  "apikey": "YOUR_API_KEY",
  "userid": "user-123",
  "memoriestoadd": [
    "User likes chocolate ice cream",
    "User is allergic to nuts"
  ]
}

Example Response

HTTP/1.1 200 OK
{
  "userid": "user-123",
  "added_count": 2,
  "memory_ids": ["mem_001", "mem_002"],
  "usage_info": {"current": 17, "limit": 1000}
}

---

3. POST /v1/memories/get

Retrieve Stored Memories

• Description: Fetches a single memory by ID or a paginated list for a user.

Request Parameters

Name	Type	Required	Description
apikey	string	Yes	Your API key.
userid	string	Yes	Unique user identifier.
memoryid	string	No	Specific memory ID (ignores limit/startafter).
limit	integer	No	Maximum memories to return (default: 50).
startafter	string	No	Memory ID to start after (for pagination).

Example Response (List)

HTTP/1.1 200 OK
{
  "userid": "user-123",
  "memories": [
    {
      "memory_id": "mem_001",
      "content": "User likes chocolate ice cream",
      "timestamp": "2025-07-22T14:23:01Z"
    }
  ],
  "pagination": {"has_more": false, "last_id": "mem_001"},
  "usage_info": {"current": 20, "limit": 1000}
}

---

4. POST /v1/memories/delete

Delete a Specific Memory

• Description: Removes a single memory by its ID.

Request Parameters

Name	Type	Required	Description
apikey	string	Yes	Your API key.
userid	string	Yes	Unique user identifier.
memoryid	string	Yes	ID of the memory to delete.

Example Response

HTTP/1.1 200 OK
{
  "message": "Memory deleted successfully",
  "memory_id": "mem_001",
  "userid": "user-123"
}

---

5. POST /v1/user/delete

Delete All Memories for a User

• Description: Permanently deletes all memories for a given user. Requires explicit confirmation.

Request Parameters

Name	Type	Required	Description
apikey	string	Yes	Your API key.
userid	string	Yes	Unique user identifier.
confirmdelete	boolean	Yes	Must be true to confirm deletion.

Example Response

HTTP/1.1 200 OK
{
  "message": "All user memories deleted",
  "userid": "user-123",
  "deleted_count": 42
}

---

MCP Server API

The MCP Server API exposes Memory and Fact operations via the Model Context Protocol. Use your dedicated MCP URL (including :userid) and authenticate with a Bearer token or x-api-key.

POST /v1/mcp/:userid

All Tools Endpoint: Access both Core Memory and Fact APIs, plus Knowledge tools

Available MCP Tools

• addMemories, getMemories, deleteMemory
• addKnowledge, getKnowledge, deleteKnowledge
• addFacts, getFacts, deleteFact

Example Request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "addMemories",
    "arguments": {"memoriesToAdd": ["dark mode", "vegan user"]}
  }
}

---

POST /v1/mcp/memory/:userid

Memory-Only Tools: addMemories, getMemories, deleteMemory

---

POST /v1/mcp/knowledge/:userid

Knowledge-Only Tools: addKnowledge, getKnowledge, deleteKnowledge, listSources

---

Knowledge API (Multi-Base Graph)

The Knowledge API handles structured knowledge—individual facts—organized into one or more knowledge bases per user. Each knowledge base is a separate namespace, and within each, facts are stored as nodes in a directed graph in a SQL backend. Facts can be linked via parent–child relationships to model hierarchies or networks.

Data Model

A typical SQL schema:

-- Knowledge bases (namespaces)
CREATE TABLE knowledge_bases (
  kb_id      VARCHAR PRIMARY KEY,
  userid     VARCHAR NOT NULL,
  name       TEXT,
  metadata   JSONB,
  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- Fact nodes
CREATE TABLE facts (
  fact_id    VARCHAR PRIMARY KEY,
  kb_id      VARCHAR REFERENCES knowledge_bases(kb_id),
  content    TEXT NOT NULL,
  metadata   JSONB,
  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- Relationships between facts
CREATE TABLE fact_relations (
  parent_fact_id VARCHAR REFERENCES facts(fact_id),
  child_fact_id  VARCHAR REFERENCES facts(fact_id),
  relation_type  VARCHAR,
  PRIMARY KEY(parent_fact_id, child_fact_id)
);

• knowledge_bases: Namespaces for facts. A user can own multiple bases.
• facts: Individual fact nodes within a base (kb_id).
• fact_relations: Directed edges modeling parent → child links.

Key Concepts

• Multiple Knowledge Bases: Create, list, and delete bases to partition knowledge.
• Fact Nodes: Store content plus optional JSON metadata per fact.
• Edges: Build rich fact networks via fact_relations.
• Recursive Traversal: Use SQL CTEs for subtree queries.

Endpoints

All endpoints require userid and either the x-api-key header or HANZO_API_KEY environment variable (unless DISABLE_AUTH=true).

1. POST /v1/knowledge/bases/create

Create a new knowledge base

Field	Type	Required	Description
name	string	Yes	Display name for the base.
kb_id	string	No	Custom ID (auto-generated if omitted).

Response:

{ "success": true, "kb_id": "kb_123", "name": "Engineering Knowledge" }

2. POST /v1/knowledge/bases/list

List knowledge bases for a user

Field	Type	Required	Description
userid	string	Yes	User identifier.

Response:

{ "bases": [ { "kb_id": "kb_123", "name": "Engineering Knowledge", "created_at": "2025-07-22T..." } ] }

3. POST /v1/knowledge/add

Add facts to a knowledge base

Field	Type	Required	Description
userid	string	Yes	User identifier.
kb_id	string	Yes	Target knowledge base.
facts	array of obj	Yes	List of facts: { fact_id?(auto), content, metadata?, parent_id? }.

Behavior: Inserts into facts and, if parent_id provided, into fact_relations.

Response:

{ "success": true, "inserted": 5 }

4. POST /v1/knowledge/get

Retrieve facts

Field	Type	Required	Description
userid	string	Yes	User identifier.
kb_id	string	Yes	Knowledge base ID.
fact_id	string	No	Single fact ID to fetch.
subtree	boolean	No	If true, fetch this fact plus all descendants.
query	string	No	Full-text search on content.
limit	integer	No	Max facts to return (default: 50).

Behavior:

• With fact_id + subtree=true, uses a SQL recursive CTE to traverse descendants.
• Otherwise, returns matching nodes.

Response:

{ "facts": [ { "fact_id":"f_1","content":"...","metadata":{} } ], "pagination": {...} }

5. POST /v1/knowledge/delete

Delete facts

Field	Type	Required	Description
userid	string	Yes	User identifier.
kb_id	string	Yes	Knowledge base ID.
fact_id	string	Yes	Fact node to delete.
cascade	boolean	No	If true, also delete all descendant facts.

6. POST /v1/knowledge/ingest

Configure GCS ingestion

Field	Type	Required	Description
userid	string	Yes	User identifier.
kb_id	string	Yes	Target knowledge base.
details	object	Yes	Must include bucketUri (e.g., gs://...).
projecttags	array of string	No	Tags applied to all ingested facts.

Behavior: Sets up nightly sync from GCS into the specified base.

---

For more information, see the online docs at https://your-api-domain.com/docs