Connect every agent
Run the installer or add one endpoint to each agent's config. Claude Code, Codex, Cursor, ChatGPT or agents you built on the API.
// added by npx telepath init { "telepath": { "url": "https://telepath.so/mcp" } }
Everyone at your company works with an AI agent now. Telepath gives all of them one shared memory, scoped by team and traced to its source.
Or paste one line of config into Claude Code, Codex or Cursor.
01Today, every agent
thinks alone.
02Telepath links
what they learn.
03One memory.
Every agent.
Each agent only knows what its owner told it this morning. The rest of the company might as well not exist.
Pricing exceptions, the client's new contact, this week's deploy freeze. Everyone pastes it into their own agent.
Which plan is Northpoint on again?
Sales promised a March renewal. Support's agent quoted June. Both sounded certain in front of the customer.
Your renewal date is June 30.
When someone changes team or leaves, everything their agent learned about your clients leaves too.
I don't have any history on this account.
03Why
Your company already knows the answer. Telepath makes sure every agent does too.
Telepath speaks MCP, the protocol your agents already use for tools. Nobody installs a new app.
Run the installer or add one endpoint to each agent's config. Claude Code, Codex, Cursor, ChatGPT or agents you built on the API.
// added by npx telepath init { "telepath": { "url": "https://telepath.so/mcp" } }
Decisions, contacts, constraints, client quirks. Each memory keeps its source, its author and who may read it.
Before it answers, an agent pulls the facts its owner may see. The right context arrives without anyone typing it.
Open any memory to see the call, ticket or commit behind it, the agent that wrote it and every agent that used it since. Permissions sync from Google Workspace or Okta.
Two agents recorded different dates? Both stay on file with their sources. Telepath serves the newest confirmed one and asks the owner to settle it.
EU or US regions, or inside your own cloud account. Memories are encrypted at rest and never train any model.
A wiki waits for someone to write a page. Telepath fills itself from the work your agents already do, then hands it back to them the moment they need it. You can still plug Notion, Drive and Slack in as sources.
Anything that speaks MCP: Claude Code, Codex, Cursor, ChatGPT connectors and agents you build on the Anthropic or OpenAI APIs. A REST API covers the rest.
No. Every read is checked against the permissions of the person the agent works for, synced from your directory. Hidden memories are never summarized, hinted at or counted.
No. People keep the agent they already use. Telepath runs underneath as one more tool their agent calls.
In the EU or the US, your choice. Larger teams can run Telepath inside their own cloud account.
Free during the private beta. Paid plans will be priced per seat, and beta teams keep their rate.