Works with
Your firm's memory, inside ChatGPT
Your firm adopted ChatGPT and it still does not know what your firm knows. Nemosyne connects the firm's own record to it.
How the connection works.
- Create your organisation and connect a source. The first companies and people appear within a minute of the first document.
- Add your organisation's MCP endpoint to ChatGPT, and sign in through OAuth as yourself.
- Ask about your own matters, deals or engagements. What comes back names its documents.
What changes against uploading files.
- Lifetime
- An upload lasts as long as the conversation. The memory outlives every thread and every person who used it.
- Permissions
- An uploaded file carries none. Nemosyne trims every answer to what the person asking is allowed to see.
- Contradiction
- Two uploaded documents that disagree produce a confident average. Nemosyne keeps both, resolves by authority and recency, and says that they disagree.
- Provenance
- An answer from an upload cites a file you pasted. An answer from the memory names the source item and can be opened.
- Effort
- Uploading is work repeated per person per thread. Reading is done once, at write time, for the whole firm.
What the connection does not do.
It does not widen anyone's access, train a model on your documents, or write anything back into your sources.
What we enforce, and the limits we publish →Questions this raises.
- Do we need an enterprise plan?
- You need a ChatGPT that can add an MCP server. The endpoint and the per-person sign-in are ours, and the same for every agent.
- Can two agents share one memory?
- Yes. The endpoint is per organisation, not per agent — Claude, ChatGPT and Cursor can all be connected to the same memory at once.
- What does it cost to connect?
- Nothing. The agent connection is never metered, and neither are reads or citations. Documents are the meter.
Connect the agent you already use.
Free for three people and about a hundred documents. No card, no expiry.
Start free →See also Claude · Cursor · Any MCP agent