Journal

How to keep a project's context across AI assistants

If you use one assistant for research, another for writing and a third for images, every new chat starts from zero. A one-page project brief fixes most of it. Template included.

, 4 minute read

Most people who do creative work with AI use more than one assistant. One is better at research, another writes in a voice you like, a third handles images. Each is good at its job, and none of them knows what the others did yesterday.

People describe the result the same way again and again. The work is spread across dozens of chats. Old material is hard to find. Every new conversation starts with ten minutes of copying and pasting to rebuild what the project is about. Long threads eventually hit a limit and have to be restarted, and the thinking in them is lost.

This guide shows a simple fix that works with any assistant: a one-page project brief that you own and bring to every chat.

Why assistants lose your context

Three things work against you.

Each assistant keeps its memory to itself. Most assistants now offer some kind of project space or memory, and the details change every few months. What one assistant learned about your project stays with that assistant, so switching tools means starting over.

Conversations get diluted. In a long thread, the brief you gave at the start competes with everything said since. The assistant may follow a decision from message 80 while forgetting the constraint from message 2.

The context lives in your head. You remember that the client hates green and the launch moved to March. Unless that is written down somewhere you can paste from, every assistant has to be told again, and some days you will forget to tell it.

The fix: a portable project brief

A project brief is one short document per project, kept somewhere you control. You paste it at the start of any chat, in any assistant, before the task. It gives every tool the same understanding of the project in a few seconds.

Copy this template into your notes app and fill it in for one live project:

Project brief template
PROJECT
Name:
One-line idea (in your words):
Who it's for:
How it should feel (three words):
It must never feel like:

DECISIONS SO FAR (newest first)
2026-10-24  Dropped the serif headline. Too close to a competitor.
2026-10-20  Launch moved to March. Shoot is now in January.

REFERENCES
- [link or description] — what you like about it

VOICE SAMPLE
One paragraph written the way this project should sound.

FOR THIS CHAT
Task:
What I want back (format, length):
Ask me before assuming anything. Don't invent facts or names.
Flag anything you're unsure about.

Keep it under about 400 words. A brief that takes a minute to read gets pasted. One that takes ten minutes gets skipped.

How to use it day to day

Start every chat with the brief

Paste the brief first, then the task. Do this even in an assistant with its own project memory. It costs a few seconds and means every tool starts from the same page.

Split by project, never by tool

Keep one brief per project rather than one per assistant. The project is the thing that lasts. Tools come and go, and you will switch between them more often than you expect.

Close each session by updating the decisions

At the end of a working session, ask the assistant: "List any decisions we made in this chat, one line each, with the reason." Read the list, keep what is true, and add it to the top of Decisions so far. This step stops old ideas from being re-suggested next week.

Restart long threads deliberately

When a thread gets long or starts drifting, ask for a short summary of where things stand, then open a fresh chat with the brief and that summary. You lose very little, and the assistant's attention is back on what matters.

Keep the voice sample honest

If the project has a voice, write the sample paragraph yourself. An assistant imitating your writing gets closer to your voice. An assistant imitating its own earlier imitation drifts further from it each time.

What this won't fix

A brief carries your context. It doesn't carry the work itself. Files, drafts and results still live in whichever chat produced them, and you are still the one moving them between tools and keeping track of what is finished. For a few small projects that is fine. Across many projects and several assistants it becomes a job of its own.

Where Pelagos comes in

That job is why we are building Pelagos. Each project gets a room that holds its context, decisions and history, so you don't have to paste a brief to remind the room what it is for. Pelago, the assistant at the centre, hands tasks to the connected AI tools you already use, where supported, and shows you what is happening, what is finished and what needs you.

Until then, the brief above works in any assistant, today. Pelagos is in private preview, and you can join the waitlist below.

Be one of the first to use Pelagos.

A private, connected workspace for turning conversations into coordinated work. We’re opening it to a small group first.

Pelagos is in private preview, Mac first. By joining, you agree we can email you about Pelagos. We won’t share your details.