Memori for OpenClaw

Memori gives OpenClaw agents a structured, long-term memory system. It automatically captures what happens and lets agents recall it on demand — so context survives across sessions without bloating the prompt.

Instead of relying solely on natural-language memory, Memori structures persistent memory from both conversation and agent trace — the agent's actions, tool results, decisions, and outcomes — so it can recall what actually happened when it matters.

The problem

OpenClaw's default memory works for simple use cases, but breaks at scale:

• Memory is stored as flat markdown files
• Context is lost due to compaction
• Important decisions and constraints disappear
• No relationships between facts
• Memory bleeds across users and projects

What Memori changes

Memori replaces flat memory with structured, scoped memory built from:

• Agent execution (tool calls, results, decisions, outcomes)

Instead of replaying history, agents retrieve exactly what they need.

How it works

Memori runs on two parallel systems:

1. Advanced augmentation

After each interaction, Memori converts raw session data into structured, reusable memories asynchronously.

• Transforms raw agent sessions into structured memory units
• Captures the agent's actions, reasoning, tool usage, responses, corrections, and failures
• Organizes into classes to enable efficient retrieval
• Generates embeddings for semantic retrieval
• Updates structured memory and the knowledge graph

This is how structured memory is continuously built and updated over time.

It runs after the agent responds and does not impact latency.

2. Agent-Controlled Intelligent Recall

Recall is explicit and initiated by the agent.

Memori separates memory creation from memory recall:

• Creation is automatic (advanced augmentation)
• Recall is intentional (agent-controlled)

Agents decide:

• When to recall
• What scope to recall from
• How much history to include

Memori does not automatically inject memory into the prompt. The agent retrieves only the context it needs, keeping token usage efficient.

Available tools:

• memori_recall — query structured memory for facts, constraints, decisions, and patterns
• memori_recall_summary — retrieve summaries and the daily brief
• memori_feedback — report on memory quality to improve the system

Quickstart

Prerequisites

• OpenClaw v2026.3.2 or later
• A Memori API key from app.memorilabs.ai
• An Entity ID to scope memory to a specific user, agent, or system
• A Project ID to scope memory to a specific project or workspace

1. Install

openclaw plugins install @memorilabs/openclaw-memori
openclaw plugins enable openclaw-memori

2. Configure

openclaw memori init \
  --api-key "YOUR_MEMORI_API_KEY" \
  --entity-id "your-app-user-id" \
  --project-id "my-project"

3. Verify

openclaw gateway restart
openclaw memori status --check

Expected:

Status: Ready

4. Test the memory loop

1. Tell the agent something durable:

   "I always use TypeScript and prefer functional patterns."

2. Start a new session and ask:

   "Write a hello world script in my preferred language."

3. Confirm the agent used memori_recall to fetch your preferences:

   [Memori] memori_recall params: {"projectId":"my-project","query":"preferred programming language"}

If it works, you now have persistent memory across sessions.

Memory model

Memory is scoped to prevent noise and ensure relevance:

• entity_id → user, agent, or system context
• project_id → project or workspace context
• session_id → specific session (requires project_id)
• date_start / date_end → time-bounded recall (defaults to all-time if omitted)
• source → type of memory (recall only)
• signal → how the memory was derived (recall only)

All timestamps are stored in UTC.

Agent behavior (read this)

Agents should:

• Retrieve a summary at the start of meaningful sessions
• Use targeted recall (not broad queries)
• Avoid recalling on every turn
• Use memory only when context is needed
• Send feedback when memory is missing or incorrect

See SKILL.md for full behavior guidelines.

Typical workflow

1. Start session → retrieve summary
2. During task → targeted recall
3. Missing context → send feedback
4. End of session → memory is captured automatically

Multi-agent ready

The plugin is fully stateless and thread-safe. You can run it across multiple agents in the same gateway without shared state or concurrency issues.

Contributing

We welcome contributions from the community! Please see our Contributing Guidelines for details on code style, standards, and submitting pull requests.

To build from source:

# Clone the repository
git clone https://github.com/memorilabs/openclaw-memori.git
cd openclaw-memori

# Install dependencies and build
npm install
npm run build

# Run formatting, linting, and type checking
npm run check

Support

• Documentation
• Discord
• Issues

License

Apache 2.0 - see LICENSE