Memory Tools

Store and retrieve information across conversations

Memory tools enable agents to persist and recall information across conversations.

Available Tools

memory_store

Store new information

memory_recall

Search and retrieve memories

memory_list

List all stored memories

memory_forget

Delete specific memories

memory_store

Store information for later retrieval.

Parameters

content · string · required

Information to remember

category · string

Category for organization: fact, preference, task, context

tags · array

Optional tags for filtering

Example

{
  "content": "User prefers TypeScript for new projects",
  "category": "preference",
  "tags": ["programming", "languages"]
}

memory_recall

Search memories using semantic similarity or keywords.

Parameters

query · string · required

Search query

category · string

Filter by category

limit · integer

Maximum results (default: 10)

Example

{
  "query": "programming language preferences",
  "category": "preference",
  "limit": 5
}

Response

Returns array of matching memories with relevance scores:

[
  {
    "content": "User prefers TypeScript for new projects",
    "category": "preference",
    "tags": ["programming", "languages"],
    "timestamp": "2026-03-03T12:00:00Z",
    "relevance": 0.95
  }
]

memory_list

List all memories, optionally filtered by category.

Parameters

category · string

Filter by category

Example

{
  "category": "fact"
}

memory_forget

Delete memories by ID or pattern.

Parameters

id · string

Memory ID to delete

pattern · string

Delete all memories matching pattern

Example

{
  "pattern": "old project"
}

Memory Categories

fact

Objective information and data

Examples:

• “API endpoint is https://api.example.com”
• “Project uses Python 3.11”

preference

User preferences and choices

Examples:

• “User prefers concise responses”
• “Avoid emojis in documentation”

task

Ongoing tasks and TODOs

Examples:

• “Need to update README with new features”
• “Pending code review for PR #123”

context

Conversation context and state

Examples:

• “Currently debugging authentication issue”
• “Working on feature branch: auth-v2”

Backend Support

Markdown (Default)

Stores memories as markdown files in ~/.zeroclaw/memory/

Features:

• Human-readable format
• Git-friendly
• No dependencies
• File-based search

SQLite

Stores memories in SQLite database with vector embeddings

Features:

• Semantic search with cosine similarity
• Fast full-text search
• Structured queries
• Requires embedding model

PostgreSQL

Enterprise-grade memory with pgvector extension

Features:

• Distributed memory across agents
• Advanced vector search
• Transaction support
• Requires PostgreSQL with pgvector

Configuration

[memory]
backend = "sqlite"  # or "markdown", "postgres"
embedding_model = "text-embedding-3-small"
max_memories = 10000

[memory.sqlite]
path = "~/.zeroclaw/memory.db"

[memory.postgres]
url = "postgresql://user:pass@localhost/zeroclaw"

Source Code

• src/tools/memory_store.rs
• src/tools/memory_recall.rs
• src/memory/

Related

• Memory Trait
• Memory Concepts
• Custom Memory Guide