MemoriaPublic betaMemory for your AI agents.
They never start from scratch again.
Memoria gives Claude Code, Codex, OpenClaw and any MCP-compatible agent a persistent, shared memory. Your preferences, your projects, your decisions: remembered from one session to the next — on your machine, and nowhere else.
$ curl -fsSL https://raw.githubusercontent.com/Primo-Studio/openclaw-memoria/memoria-v1/scripts/install-memoria.sh | shRequirements: Node.js 22 LTS (free, nodejs.org) · macOS (Apple Silicon & Intel)
Every conversation starts from zero.
You explain your project, the agent works, the session ends — and everything is gone.
The next day, you re-explain the same architecture, the same conventions, the same decisions. Your agents forget everything; you repeat everything.
Up and running in three steps.
Install
A single command in the Terminal. The script takes care of the rest: tooling, automatic startup at login, opening the interface.
Pick your intelligence engine
Ollama (recommended: 100% local and free), LM Studio, or your own OpenAI, OpenRouter or Anthropic API key. Guided onboarding walks you through it — and if no engine is configured, Memoria says so clearly.
Connect your agents
Memoria detects the agents installed on your machine. One-click connection and, if you want, import of their existing memories — every import goes through a review screen you approve.
A complete memory. Under your control.
100% local
Everything lives on your machine: the database, the search, the interface (127.0.0.1, token-protected). No cloud, no account.
Private memory per agent
Each agent has its own memory. What one learns doesn't leak to the others — unless you decide otherwise.
Sharing governed by you
Facts about you, your company and your projects can be shared across agents. You decide what, and with whom.
Import existing memories
Bring in your Claude Code and Codex conversations, and your OpenClaw memory. You approve every import on the review screen before anything is saved.
Powerful search
Keywords + semantic search across your themes, their relations, your procedures and your contacts. Find a three-month-old decision in two seconds.
Secrets in the Keychain
API keys and passwords go into the native Keychain, never in plain text. Secrets are automatically redacted before anything is stored.
Pause & full control
Incognito mode (kill switch), permanent deletion, audit log, storage you can move to a USB drive. Your memory belongs to you.
One-click updates
An Update button built into the interface. No reinstalling, no trips back to the Terminal.
Coming soon: encrypted sync between your machines (hub-and-spoke). Already in the code, currently being field-tested.
Your agents, ready to go.
Memoria automatically detects the agents installed on your machine and connects to them through the open MCP protocol.
Automatic detection, one-click connection, import of existing conversations.
Automatic detection, one-click connection, import of existing conversations.
Dedicated adapter, with import of your existing OpenClaw memory.
Memoria speaks MCP: any agent compatible with the protocol can connect.
Your memories never leave your machine.
Memoria runs entirely locally. The interface is only reachable from your machine (127.0.0.1, token-protected), and the code is public: you can verify every one of these claims yourself.
Same idea. A different choice.
Cloud memory layers for AI agents exist: hosted APIs, subscription pricing, your data on their servers. Memoria goes the other way — the full memory layer, installed on your machine.
Cloud memory layers are good tools for hosted products at scale. Memoria answers a different question: your memory, yours, at home.
Apache-2.0. The code is public.
Memoria is released under the Apache-2.0 license. Read the code, contribute, report a bug: the project is built in the open, and your beta feedback genuinely matters.
The questions we get asked.
Is it really free?
Yes. Memoria is open source (Apache-2.0 license) and free. No subscription, no paid tier, no locked features. With Ollama as your engine, the whole thing runs without spending a cent.
Where does my data go?
Nowhere. Everything is stored and processed on your machine: no cloud, no account, no telemetry. The only exception is if you yourself choose an API-key engine (OpenAI, OpenRouter, Anthropic): analysis requests then go through that provider. With Ollama or LM Studio, nothing ever leaves.
Which agents are compatible?
Claude Code, Codex and OpenClaw are detected automatically and connected in one click, with import of their existing memories if you want it. And since Memoria speaks MCP, any agent compatible with the protocol can connect.
Do I need an API key?
No. The recommended engine is Ollama: 100% local and free. You can also use LM Studio, or an OpenAI, OpenRouter or Anthropic API key if you prefer. And as long as no engine is configured, Memoria tells you so clearly instead of pretending.
Does it work on Windows or Linux?
macOS today (Apple Silicon & Intel). Linux should work but hasn't been tested yet. Windows isn't supported yet. The project is open source: contributions are welcome.
How do I update?
From the interface: a built-in Update button downloads and installs the new version. No need to go back to the Terminal.
What if I want to erase everything?
Permanent deletion is built in: you can erase one agent's memory, or all of Memoria, straight from the interface. And since everything is local, deleting really means deleting — no copy exists anywhere else.
Give your agents a memory.
In one command.
Open source, free, 100% local. The public beta is open: install it, test it, and tell us what you think.
$ curl -fsSL https://raw.githubusercontent.com/Primo-Studio/openclaw-memoria/memoria-v1/scripts/install-memoria.sh | sh