Portable AI memory
One memory.
Across the AI apps you choose.
The Kelven MCP Gateway connects compatible AI apps to the memory you keep in Kelven. Instead of rebuilding your preferences, projects and context in every app, each one can ask Kelven for the small piece it needs — with your permission.
Your context, not another chatbot
Keep the memory. Change the interface.
ChatGPT and other compatible apps can use what you have already taught Kelven. The connection is read-only: it can ask for relevant context, but it cannot add, edit or delete your memory.
Search memory
Find the few records relevant to a question. There is no capability that returns everything.
Get the context a task needs
The signature capability: state the task, receive the minimum context for it — not a profile.
Search projects
Find a project by name or description.
Get project context
One project's summary and the records attached to it, nothing further.
Read preferences
How you like answers written, formatted and delivered.
Read privacy rules
The limits Kelven enforces, so an assistant knows what it may not ask for.
When an app asks Kelven
Kelven returns context, not your whole memory.
These protections apply to the memory and context Kelven returns through the connection. They do not change or conceal the message you typed directly into the other AI app.
- 01 · Authorization
You approved this application, and only the permissions shown.
- 02 · Relevance
Only records related to the stated task are considered.
- 03 · Sensitivity filter
Local-only records stop here. So does anything you excluded.
- 04 · Minimization
What remains is trimmed to the few records the task needs.
- 05 · Tokenization
Names, companies, amounts and addresses become typed placeholders.
- 06 · Result
The assistant receives context plus a plain summary of what was withheld.
Local-only memory
Some memory never leaves your device. Ever.
A record you mark local-only cannot be returned through the Gateway. Not tokenized, not summarised, not "just this once". A connected application is told plainly: requested memory is stored locally and is not available to connected applications.
Kelven will never quietly upload local memory to make a connection work.
The important boundary
MCP carries memory. It does not intercept your prompt.
If you type a client's real name into ChatGPT or another AI app, that app receives it before it can ask Kelven for anything. The MCP connection cannot inspect, rewrite or sanitize that message.
MCP protects only the context Kelven supplies from your memory. To protect the prompt itself before it reaches another provider, use Kelven's Browser Shield or Privacy Gateway instead. We also only claim compatibility with apps that support remote MCP.
Compatibility matrix
Tested honestly, labelled honestly.
| Interface | Status | How it connects |
|---|---|---|
| Kelven app | Complete experience | Chat, routing, memory, privacy, projects |
| ChatGPT | Supported MCP client | Custom connector, developer mode |
| Other MCP clients | Compatible | Any client supporting remote MCP over HTTP |
Nothing is marked Verified yet. Statuses change only after real end-to-end testing.
What an MCP server does
How a server exposes capabilities, and why narrow beats broad.
Read the explainerYour AI, your context. Start with Kelven today.
Put powerful AI to work with memory you control and protection before cloud requests.