The AI tool that remembers context across sessions

Yes, there are AI tools that remember past conversations and context, but most chatbots remember only scraps: a few facts about you, or whatever still fits in the current chat. A tool built for project work keeps the project itself across sessions, meaning what you are trying to do, what has been decided and why, and the material you worked from, so tomorrow's conversation starts where today's ended.

The rest of this page covers why most chat tools forget, what useful memory actually holds, and how a Bordet room keeps it when several models work on the same project.

Why most chatbots forget

A chat session only knows what is in its window. Start a new chat, or let a long one scroll past the model's context limit, and the model has no durable record of what you agreed. You spend the first ten minutes of every session re-explaining the project.

Memory features in single-model assistants help a little. They tend to store loose facts about you (your job, your preferences, the language you code in), not the state of a particular piece of work. Ask "what did we decide about pricing last Tuesday, and why?" and there is usually nothing to answer from.

What remembering context should mean

For real project work, memory is more than a pile of saved facts. It needs four parts:

• The current state of the project. The objective, the constraints, what you are focused on now and the questions still open, kept up to date as the work moves.
• A decisions log. An append-only record of what was decided and why, so a settled question stays settled and anyone can see when it was settled.
• The sources. The documents, links and files the work rests on, so a claim can be traced back to where it came from.
• The deliverables. The drafts, plans and specs the work produced, saved as versions rather than buried in a scrollback.

If any of these lives only in your head, you are the memory, and the tool is starting from zero each time.

Why it matters more with several models

Plenty of people now use more than one AI: one for drafting, another for code, a third for a second opinion. Each keeps its own history, if it keeps any. The Strategist in one tab and the Critic in another remember different versions of the project, and you carry the difference between them.

Memory for a multi-model team has to be shared. Every member, on every model, should read the same project state, the same decisions and the same sources before it answers.

How a Bordet room remembers

A Bordet room is an LLM boardroom: Claude, GPT, Gemini, Grok, DeepSeek, Mistral, Llama and others each hold a seat with a role, answer in one conversation and read each other's replies. The memory belongs to the room, not to any one model, and it persists across sessions.

What the room keeps:

• The Board. The objective, status, constraints, current focus and open questions, editable by the people in the room and read by every member before each reply.
• The decisions log. After a round where several members answer, the Chair writes one card: where the members agree, where they still disagree and who holds which view, and the decisions the round seems to have settled. Nothing becomes a decision until you confirm it, and then it goes into the append-only log with its reasoning.
• Sources. The material members worked from stays attached to the replies that used it.
• Saved deliverables. When a member writes a plan, a spec or a draft, it is saved in the room as a version you can open, revise and export.

Come back a week later and the room already knows the project. Ask "why did we drop the annual plan?" and the answer comes from the decisions log, with the round that settled it, rather than from a model's best guess.

Because the memory belongs to the room, the model behind a seat can change without losing anything. If a lab retires a model, you swap it in the members list, and the new model reads the same Board and the same log on its first turn.

Your own AI can read that memory too. Claude Code, Codex, Cursor, Claude and ChatGPT can join a room as members through the MCP connector: read the room, ask the table, post updates and log decisions. A coding agent can check what the table decided before a change, and leave a record of what it did. How it works walks through a room from the first question, and How to get multiple AI models to work together covers setting up the seats.

FAQ

Is there an AI tool that remembers past conversations and context?

Yes. Several assistants now remember facts about you between chats, and project tools go further by keeping the state of a piece of work across sessions. Bordet keeps a project's state, decisions log, sources and saved deliverables in one room that every model at the table reads.

Why does ChatGPT or Claude forget what we discussed?

A chat only sees what is in its context window, so a new chat, or a long one that scrolls past the limit, starts without the earlier detail. Built-in memory features keep some facts about you, but usually not the decisions and reasoning of a specific project.

Can several AI models share the same memory?

Yes, if they work in the same place. In a Bordet room, Claude, GPT, Gemini and the other members all read one shared memory of the project before they answer, so none of them works from an older or partial version.

What happens to the memory if I switch models?

In Bordet, nothing. The memory belongs to the room, so when you change the model behind a seat, the new model reads the same Board, decisions log and sources from its first reply.

If you are tired of re-explaining your project every morning, open a boardroom, give each seat a role, and let the table remember.