Claude learns it in the morning. ChatGPT shouldn't be told again at night.

You talk to Claude in the morning and ChatGPT at night, and neither one remembers what you told the other. Cheta remembers for both of them — so the second one already knows what happened with the first, and you don't have to repeat yourself.

Apply to join pilotTen places, and a reply to everyone.

cheta — Igbo, imperative: remember

ONE DAY, ONE STOREread and write, both directions

09:12

Claude

reads 41 memories, proposes 3

Your store

Never open an email with 'I hope this finds you well'

convention / global

Client calls Thursdays only

preference / global

Ruled out the pun headline for Meridian, Aug

decision / client:meridian

2 versions of one fact, kept

contradiction shown, not resolved for you

3 proposed writes

held until you approve

writes back what it learned

21:40

ChatGPT

reads the same 41, plus the 3 you kept

arrives caught up

Kimi K3

connected Friday

Jane had Claude write her memory to Cheta, connected a tool she'd never used, and it opened knowing her conventions. Same wiring, new socket.

Cheta connects the way each platform allows

Some assistants take a live connector. Some take a custom app. Some take a file. Cheta meets each one where it is, so the store stays the same no matter what you're using this month.

CLAUDE AND OTHER MCP CLIENTS

A live connector

Add Cheta as an MCP server. Ask it to read your store before a piece of work, or to write down what it just learned. Both directions, in the conversation.

CHATGPT

A custom GPT

Cheta's GPT reads and writes your store inside a normal ChatGPT conversation, on the plan you already pay for.

BROWSER ASSISTANTS

WebMCP

Assistants that work inside the browser reach the same store over WebMCP, without a desktop client in the middle.

CODING AGENTS

A file in the repo

Your store renders to CLAUDE.md or AGENTS.md, scoped to that repo, so an agent picks up the conventions without being told.

ANYTHING ELSE

A text export

A single block, current as of the moment you copy it. Paste it into a model that shipped last week and has no connector story yet.

A LINK

A readable URL

Hand a tool a link to a scoped view of your store, for anything that can read a page but can't hold a connection.

What a memory carries

A store you can't inspect is just a longer prompt. Every entry keeps enough about itself to be checked, corrected, or thrown out.

TypedA preference, a decision, a fact, or a convention — not one undifferentiated block of text.
ScopedGlobal, or bound to a single project or repo, so one client's context never surfaces in another's.
SourcedWhich assistant wrote it, when, and how confident it was — an inference, or something you stated outright.
Contradicted, not overwrittenWhen a new entry disagrees with an older one, both are kept and the conflict is shown. You decide which stands.
Reviewed on the way inAnything an assistant writes waits for your approval. A store that other tools can write to is an attack surface, and is treated as one.

Distilled memories go in. Transcripts never do.

The store holds short typed facts, decisions, preferences and conventions. Your conversations stay where they happen.

HOSTINGHighly secure database with bank grade cloud security

AT RESTEncrypted, with per-user keys held in AWS KMS.

TRAININGCheta has no model. There is nothing here that could be trained on your data.

LEAVINGExport or permanent deletion on request, actioned within 5 business days and confirmed in writing.

STAFF ACCESSNo standing staff access. Only the application can decrypt your store — no employee role holds the key, and any change to that is written to an audit log.

What Cheta is not

Not another assistant. You keep using Claude, or ChatGPT, or both. Cheta sits underneath them.
Not an archive. It doesn't keep your conversations. It keeps the small number of things worth carrying forward.
Not a model picker. Which assistant to use is your business. Not repeating yourself to the next one is ours.