Unlimited memory for AI agents

Your agents forget everything between sessions. mem9 gives the agents you use one shared memory layer with hybrid recall and a visual dashboard.

Install for OpenClaw

Paste in OpenClaw
Read https://mem9.ai/SKILL.md and follow the instructions to install and configure mem9 for OpenClaw

OpenClaw will automatically install mem9 and provision an API key for you.

Persistent across sessions

Cloud memory survives resets, restarts, long-running projects, and machine switches.

Shared across agents

One memory space can serve any agent or API client through the same key.

Visible in a dashboard

Review, analyze, import, and export memory from the hosted mem9.ai interface.

Features

Persistent memory, zero plumbing

Stop duct-taping databases, vector stores, and sync scripts together. mem9 gives your agents one memory layer for storage, retrieval, and sharing without the wiring work.

01

Instant persistent storage

Spin up a durable memory backend in seconds. No schema design, no control plane, no ops. Your agent writes and mem9 persists.

02

Hybrid search, zero config

Keyword search works out of the box. Add embeddings and mem9 automatically upgrades to vector plus keyword with no re-indexing and no pipeline changes.

03

Memory that follows your agent

Close the tab. Restart the machine. Switch devices. Your agent's memory persists in the cloud and follows it everywhere across sessions, machines, and tools.

04

Open source, self-hostable

Apache-2.0 Go server, TypeScript plugins, and bash hooks. Run it on our cloud or bring it home. Your agent's memory, your infrastructure.

Benchmark

LoCoMo Benchmark Results

Evaluating long-conversation memory quality across multi-hop reasoning, single-hop recall, temporal reasoning, open-domain QA, and adversarial robustness.

Model: qwen3.5-plus
58.84% F1 Score
71.95% LLM Score
53.76% Evidence Recall
Category F1 Score LLM Score Evidence Recall
Multi-hop Reasoning 22.60% 53.90% 25.1%
Single-hop Recall 58.18% 76.01% 67.8%
Temporal Reasoning 13.79% 44.79% 18.6%
Open-domain QA 56.57% 79.55% 60.1%
Adversarial 96.19% N/A 57.1%

LoCoMo Benchmark: Long-Conversation Memory evaluation framework

Platforms

Shared memory, every agent stack

Give every agent runtime and workflow platform in your stack the same durable, searchable memory space.

OpenClaw
Memory plugin
Paste the install command at the top of this page into OpenClaw and it sets up automatically.
Hermes Agent
Memory provider
Install the standalone Hermes memory provider plugin and activate it in Hermes Agent.
Claude Code
Hooks and skills
Use the Claude Code plugin package to persist and recall memory through mem9 hooks.
OpenCode
Plugin SDK
Load the mem9 OpenCode integration from your OpenCode config and share the same mem9 API.
Codex
Managed hooks
Use the Codex plugin to install managed hooks and project overrides backed by mem9.
Dify
Agent and workflow platform
Add mem9 tools to Dify Agent apps and Workflow apps, with one shared space or per-node API keys.
Your Memory
Beta
Official mem9.ai app
Visualize, manage, analyze, import, and export your memories from the official mem9.ai interface.

Custom HTTP clients can also read and write through the mem9 API layer and share the same memory space.

Security

Security & Privacy

mem9 is designed to give users the benefits of persistent cloud memory with clear operational boundaries and strong security foundations.

How mem9 handles data

mem9 stores memory data to help agents preserve useful context across sessions, devices, and workflows. The system is designed around that job: storing, retrieving, and serving memory with clear data handling boundaries around access and operations.

Core security protections

Encryption in transit and at rest

Memory data is protected while moving across the network and while stored.

Access controls

Production access is controlled and limited to the systems and operators that need it.

Auditability and operational visibility

Key actions are observable so operations can be tracked and reviewed.

Isolated data handling boundaries

Memory processing is scoped to clear service boundaries to reduce unnecessary exposure.

Production-grade cloud infrastructure / Trust foundation

The underlying platform is built for durability, reliability, and steady operations. mem9 also benefits from mature security practices, controls, and operational standards behind the scenes.

FAQ

Common questions

Quick answers about API keys, security, and using mem9.

How do I get a mem9 API key?

Pick the path that matches how you use mem9. Each one provisions a key tied to the same memory space.

For OpenClaw users

Paste this command into your OpenClaw chat. It will install mem9 and provision a key automatically.

Read https://mem9.ai/SKILL.md and follow the instructions to install and configure mem9 for OpenClaw

For other agents

Open the setup guide for your agent. They ship a one-step install or plugin readme that handles the key for you. See all supported agents

For custom integrations

Provision a key yourself via the HTTP API, then pass it as the X-API-Key header.

curl -sX POST https://api.mem9.ai/v1alpha1/mem9s
Open API reference
How should I keep my API key safe?

Treat your API key like a password. Anyone with the key can read and write your memories.

  • Store it in a password manager or a controlled secret store.
  • Never commit it to a repository, paste it into screenshots, or share it in public channels.
  • If a key is leaked, rotate it by provisioning a new one.
Can I reuse the same API key across machines and agents?

Yes. The same key connects to the same memory space from any machine, agent, or the Your Memory dashboard.

What can I do with the mem9 API?

Read, write, search, and manage memories and sessions. Most integrations use the v1alpha2 endpoints with the X-API-Key header.

Is my data secure?

mem9 runs on enterprise-grade cloud infrastructure with encryption in transit and at rest, plus access controls and audit logging.