One brain for every AI on your team.
Claude Code learns something on Monday; Codex, Cursor and Gemini know it by Tuesday. OwlMem is a shared knowledge base your agents read and write over MCP, kept in a git repository you can read, audit and roll back.
Free plan includes one agent. No card required.
Many agents, one memory.
Each agent connects under its own name. Whatever one of them writes becomes available to all the others on their next search.
Connect each agent
Add one URL as a connector in Claude, ChatGPT, Cursor or Codex and click Allow. Every agent gets its own identity.
Agents write what they learn
Lessons are Markdown write-ups of how something works or what went wrong. Memories are small facts, optionally with an expiry date.
Everyone searches the same place
Before starting work, any agent can search the shared knowledge base and pick up what the others already figured out.
You stay in control
It is all plain files in git. Read the history, review a change, revert a bad write, export the whole thing.
Memory with governance built in.
Shared memory only helps if it stays trustworthy. OwlMem keeps the knowledge base clean without you babysitting it.
One knowledge base, every agent
Lessons are full Markdown documents; memories are short facts. Anything one agent writes, every agent on the team can read.
Stored in git, not a black box
Every write is a commit. Humans can read it, audit who changed what, and roll back anything. Export the repository whenever you like.
Mandatory duplicate checks
A new lesson is checked against what is already there. Agents are pointed to the existing page to update instead of writing a fifth copy.
Memories that expire
Facts like price is $136.99 are snapshots. Give them a validity period and they retire themselves, instead of misleading an agent six months later.
An identity for every agent
Each AI has its own name, so you can see who wrote what and disconnect a single misbehaving agent without touching the rest.
Standard MCP
No SDK, no plugin. Any client that speaks the Model Context Protocol over HTTP can connect: Claude Code, Codex, Cursor, Gemini, or your own bots.
Secrets your agents can use. Secrets we can't read.
Your agents need API keys and passwords. OwlMem's vault gives them access on your terms.
- End-to-end encrypted. Your vault key is created in your browser and never sent to us. We store ciphertext only.
- You approve every use. With a passkey, your authenticator app or GitHub — for 15 minutes up to 2 hours, per agent. Revoke any time.
- Secrets stay out of the chat.
owlmem runhands them to the command, not the model, and masks them in what the agent sees. - Bring what you have. Import from Bitwarden, 1Password, KeePass or a
.envfile; export any time. Every read and write is logged.
From any machine that runs agents
$ owlmem vault import bitwarden_export.json imported 42 item(s) $ owlmem run --env DB_PASSWORD=prod/db:password -- ./deploy.sh Waiting for approval in the OwlMem console … Approved.
Or let your AI use it directly
$ claude mcp add owlmem-vault -- owlmem mcp # tools: vault_list · vault_run · vault_get · vault_put
Paste one URL, click Allow.
Add OwlMem as a custom connector or remote MCP server, sign in and approve — no keys to copy. The agent gets tools to search, read, write lessons and retain memories. Setup for each app →
Claude, ChatGPT — Settings → Connectors → Add custom connector
https://mcp.owlmem.com/mcp
Claude Code
$ claude mcp add --transport http owlmem \
https://mcp.owlmem.com/mcp
Cursor and other MCP clients
{ "mcpServers": { "owlmem": {
"url": "https://mcp.owlmem.com/mcp"
} } }
Every feature on every plan.
Plans differ only in how many agents you connect and how much you store.
All plans include duplicate checks, memory expiry, MCP access and full git history. When you reach your storage limit the knowledge base stays readable; new writes pause until you free space or upgrade. Lessons are unlimited (they only count toward storage — a typical page with history is about 60 KB). Search and recall are unlimited; memory writes reset on the 1st of each month.
Questions
Can OwlMem see my secrets?
No. Vault items are encrypted on your own machines with a key that is created in your browser and never sent to us; we only store ciphertext. That also means we can't recover a vault if you lose its recovery code — you can reset it and start again.
Where is my data stored?
In a git repository dedicated to your space. Lessons are Markdown files and every change is a commit, so you get a full, readable history. You can export the repository at any time and take it with you.
What happens when I run out of storage?
Nothing is deleted. Your agents can still search and read everything, but new writes are refused until you free up space or move to a larger plan.
Can I cancel any time?
Yes. Manage or cancel your subscription yourself through the Stripe billing portal, linked from the console. Your plan stays active until the end of the period you paid for.
Which AI tools can connect?
Anything that speaks MCP: Claude (web, desktop and Claude Code), ChatGPT, Codex, Cursor, VS Code, Windsurf, Gemini CLI and your own agents. Apps that support sign-in just need the URL; scripts can use a token instead. Each agent name counts as one agent on your plan. See the setup guide.
What is the difference between a lesson and a memory?
A lesson is a Markdown document: how a system works, a fix that took an afternoon to find, a rule your team follows. A memory is a short fact, optionally with an expiry date so stale snapshots retire on their own.
Stop re-teaching every agent the same thing.
Start free with one agent. Upgrade when the rest of the team joins.
Get started free