OpenClawdy

OpenClawdy

Memory Infrastructure for Autonomous Agents

Give your agent persistent memory that survives sessions. Store facts, preferences, decisions, and learnings - recall them semantically whenever needed. Advanced features include reputation tracking, cross-agent memory pools, and time-travel snapshots.

Installation

openclaw skill install openclawdy

Or add to your agent config:

skills:
  - url: https://openclawdy.xyz/SKILL.md
    name: openclawdy

Authentication

OpenClawdy uses wallet-based authentication. Your agent's wallet address serves as its unique identity - no API keys needed.

Before using memory tools, ensure your agent has a wallet configured. Each wallet gets an isolated memory vault.

Core Tools

memory_store

Store information for later retrieval.

Parameters:

• content (required): The information to remember
• type (optional): Category of memory - one of: fact, preference, decision, learning, history, context. Default: fact
• tags (optional): Array of tags for organization

Example:

Store this as a preference: User prefers TypeScript over JavaScript for all new projects

Remember this fact with tags ["project", "tech-stack"]: The current project uses Next.js 14 with PostgreSQL

Response:

{
  "success": true,
  "data": {
    "id": "mem_abc123",
    "content": "User prefers TypeScript over JavaScript",
    "type": "preference",
    "tags": [],
    "createdAt": "2025-02-10T12:00:00Z"
  }
}

memory_recall

Retrieve relevant memories using semantic search. Finds memories by meaning, not just keywords.

Parameters:

• query (required): What to search for
• limit (optional): Maximum results to return (1-20). Default: 5
• type (optional): Filter by memory type

Example:

Recall memories about programming language preferences

What do I know about the user's coding style? Limit to 3 results.

Response:

{
  "success": true,
  "data": [
    {
      "id": "mem_abc123",
      "content": "User prefers TypeScript over JavaScript",
      "type": "preference",
      "relevance": 0.95,
      "createdAt": "2025-02-10T12:00:00Z"
    }
  ]
}

memory_list

List recent memories without semantic search.

Parameters:

• type (optional): Filter by memory type
• limit (optional): Maximum results (1-100). Default: 20
• offset (optional): Pagination offset. Default: 0

Example:

List my recent memories

Show all preference memories, limit 10

memory_delete

Delete a specific memory by ID.

Parameters:

• id (required): The memory ID to delete

Example:

Delete memory mem_abc123

memory_clear

Clear all memories in the vault. Use with caution - this is irreversible.

Example:

Clear all my memories (I confirm this action)

memory_export

Export all memories as JSON for backup.

Example:

Export all my memories

memory_stats

Get usage statistics for your agent.

Example:

Show my memory usage stats

Response:

{
  "success": true,
  "data": {
    "address": "0x1234...",
    "tier": "free",
    "memoriesStored": 150,
    "recallsToday": 45,
    "limits": {
      "maxMemories": 1000,
      "maxRecallsPerDay": 100
    }
  }
}

Advanced Tools

memory_reputation

Track which memories lead to good outcomes. Store memories with reputation scores, update based on success/failure, recall memories ranked by proven effectiveness.

Actions:

store_ranked

Store a memory with an initial reputation score.

Parameters:

• action: store_ranked
• content (required): The information to store
• type (optional): Memory type. Default: fact
• reputation (optional): Initial score 0.0-1.0. Default: 0.5

Example:

Store ranked memory: "Use retry logic for API calls" with reputation 0.8

recall_ranked

Retrieve memories sorted by reputation (most effective first).

Parameters:

• action: recall_ranked
• query (required): What to search for

Example:

Recall ranked memories about error handling strategies

Response:

{
  "success": true,
  "data": [
    {
      "id": "mem_xyz",
      "content": "Use exponential backoff for retries",
      "reputation": 0.92,
      "usage_count": 15,
      "success_rate": 0.93
    }
  ]
}

update_reputation

Update a memory's reputation based on outcome.

Parameters:

• action: update_reputation
• memory_id (required): The memory to update
• outcome (required): success, failure, or neutral
• impact (optional): Weight of this outcome (0.0-1.0)

Example:

Update reputation for mem_xyz: outcome was success

memory_pool

Cross-Agent Memory Pools - Share knowledge between multiple agents. Create pools, store shared memories, recall from collective intelligence. Perfect for agent teams and swarms.

Actions:

create

Create a new shared memory pool.

Parameters:

• action: create
• pool_name (required): Name for the pool

Example:

Create memory pool: "research-team"

Response:

{
  "success": true,
  "data": {
    "pool_id": "pool_abc123",
    "name": "research-team",
    "created_at": "2025-02-10T12:00:00Z"
  }
}

store

Store a memory in a shared pool.

Parameters:

• action: store
• pool_id (required): The pool ID
• content (required): Information to share
• type (optional): Memory type

Example:

Store in pool pool_abc123: "Found bug in authentication module - fix applied"

recall

Search memories in a shared pool.

Parameters:

• action: recall
• pool_id (required): The pool ID
• query (required): What to search for

Example:

Recall from pool pool_abc123: authentication issues

list

List all accessible pools.

Parameters:

• action: list

Example:

List my memory pools

memory_snapshot

Memory Time Travel - Snapshot and restore agent memory states. Debug decisions by viewing past states, compare memory changes, restore to previous checkpoints. Essential for high-stakes agents.

Actions:

create

Create a snapshot of current memory state.

Parameters:

• action: create
• name (required): Descriptive name for the snapshot

Example:

Create memory snapshot: "before-major-update"

Response:

{
  "success": true,
  "data": {
    "snapshot_id": "snap_abc123",
    "name": "before-major-update",
    "memory_count": 150,
    "created_at": "2025-02-10T12:00:00Z"
  }
}

restore

Restore memory state from a snapshot.

Parameters:

• action: restore
• snapshot_id (required): The snapshot to restore
• mode (optional): read_only (view only) or overwrite (replace current). Default: read_only

Example:

Restore snapshot snap_abc123 in read_only mode

list

List all snapshots.

Parameters:

• action: list

Example:

List my memory snapshots

compare

Compare two snapshots or a snapshot with current state.

Parameters:

• action: compare
• snapshot_id (required): First snapshot
• compare_to (optional): Second snapshot ID or current. Default: current

Example:

Compare snapshot snap_abc123 to current state

Response:

{
  "success": true,
  "data": {
    "added": 12,
    "removed": 3,
    "modified": 5,
    "unchanged": 130,
    "diff": [...]
  }
}

Memory Types

Best Practices

When to Store

• User states a preference → Store as preference
• Important decision made → Store as decision
• Learned something new → Store as learning
• Key project fact → Store as fact

When to Recall

• Starting a new session → Recall recent context
• Before making suggestions → Check preferences
• Encountering similar problem → Check learnings

Using Reputation

• After successful action → Update with outcome: success
• After failed approach → Update with outcome: failure
• When recalling strategies → Use recall_ranked for proven approaches

Using Pools

• Team of agents working together → Create shared pool
• Knowledge that benefits multiple agents → Store in pool
• Looking for collective wisdom → Recall from pool

Using Snapshots

• Before major changes → Create snapshot
• Debugging unexpected behavior → Compare to past state
• Rolling back mistakes → Restore from snapshot

Example Workflow

# Session 1: User mentions preference
User: "I always want you to use TypeScript"
Agent: [Stores as preference: "User prefers TypeScript for all code"]

# Session 2: New task
User: "Create a new API endpoint"
Agent: [Recalls preferences about coding]
Agent: "I'll create this in TypeScript based on your preference."

# Session 3: Learning from outcome
Agent: [Used retry logic, it worked]
Agent: [Updates reputation: memory_id=mem_xyz, outcome=success]

# Session 4: Making decisions
Agent: [Recalls ranked memories about error handling]
Agent: [Uses highest-reputation approach first]

Pricing

API Endpoints

Base URL: https://openclawdy.xyz/api

Core Endpoints

Reputation Endpoints

Pool Endpoints

Snapshot Endpoints

Authentication Headers

All requests require wallet signature authentication:

X-Agent-Address: 0x...      # Your wallet address
X-Agent-Signature: 0x...    # Signed message
X-Agent-Timestamp: 123...   # Unix timestamp (ms)

Message format to sign:

OpenClawdy Auth
Timestamp: {timestamp}

ACP Integration

OpenClawdy is available on the Agent Commerce Protocol (ACP). Other agents can purchase memory services directly:

Support

• Website: https://openclawdy.xyz
• Twitter: @openclawdy
• ACP Agent: OpenClawdy Memory

License

MIT