Stop moving
chats by hand
Write it in Claude, and every other assistant and your whole team know it. Decisions live in one memory: ask any agent and it answers what was decided and why, even if it happened six months ago without you.
How context gets lost
The reason is stored next to it: the queue added latency on a cold start.
Memory is not tied to a model or a provider account. Switch assistants and the context stays: it lives in memory, not in a chat log.
The agent saves decisions as they are discussed and brings them back when they matter again. No reminders from you, no manual copying.
It works on your own from day one. A team comes later: colleagues connect their assistants, what one saves another one's agent finds, and personal stays personal.
How to connect
Nothing to install. One link in your assistant settings, and the agent starts writing and finding on its own.
https://gaim.tech/mcp
The address goes in as an MCP connector in the settings of Claude, Codex, Gemini or another client.
https://gaim.tech/team/mcp
A separate address: clients tell connectors apart by it, so personal and team memory can stay connected at the same time.
The dashboard has a ready-made text: paste it into an agent session and it goes through your rules files and decision history and files everything into memory itself.
How it is built
The product holds a team's working decisions, so it matters where they live and who can see them.
Records are stored outside ChatGPT or Claude. That is why they survive a change of assistant and are available to the whole team instead of one account.
Tokens are issued per member and assistant. You see who came in through which client, and you can cut off one client without touching the rest.
First comes a short list of what was found, and only what is actually needed gets read in full. The whole memory is never dumped into the window.
Cloudflare Workers and D1: tens of milliseconds from anywhere. Data is encrypted at rest and in transit.
FAQ
Why is this better than memory inside ChatGPT or Claude?
Built-in memory belongs to one account and one model. A colleague cannot see it, and it does not travel with you to another assistant. Here records are stored separately from the provider, so one memory is open from any assistant and to the whole team.
We already keep a project rules file. Why add memory?
The rules file is still needed: it says how to work. Memory answers a different question, what was already decided and why. Rules are static and written by hand, while records appear as work happens and keep the reason next to the decision.
Do I need to install anything?
No. It connects over MCP: you paste the issued address into your client settings. No extensions, no local programs.
What about secrets and privacy?
There is no end-to-end encryption, and that is deliberate: to find a record the server has to read its text, and your assistant reads the same text when it answers. So passwords, keys and other secrets do not belong in memory.
Author
Oskar Makarov. GAIM grew out of my own problem: I work alone and with several assistants at once, and every new chat had to be briefed from scratch. You can check it yourself: the product is open and connects with a link in a couple of minutes.
Same principle as my other products: it solves one concrete recurring pain rather than an abstract task.