Answers
We are switching AI tools. Do we lose what the team already decided?
Not if the decisions were never inside the tool. What a team loses in a migration is whatever it kept in a vendor's own memory, its settings and its per-tool configuration files — those belong to that tool, and no export makes them arrive intact somewhere else. Arroway holds decisions in a different place: a server your tools connect to. The new assistant reads the same sanctioned set the old one read, on its first day, so the switch costs you configuration instead of context.
Last updated September 3, 2026
What travels, and what was never portable
Three different things get called "our AI setup", and they do not behave the same way in a migration. The wiring of a tool — its permissions, its own settings, how it is installed — does not travel, and should not: it belongs to the tool. Memory built into a tool lives inside that tool's account and scope, so a change of vendor is a change of account. What the team decided — how you name things, which client rule is in force, what was ruled out and why — is not any tool's property. Losing it in a vendor migration is an accident of where it happened to be kept.
The decision belongs to the project, not to the assistant
Each decision is written down as the work happens, approved by a person, and stored against the project. Reading it is a connection, not an installation: the assistant asks the server when it starts working and gets the current set. Change the assistant and the set is unchanged — it was never a copy living inside the old tool, so there is nothing to migrate and nothing to reconcile afterwards.
Running both tools during the move, without a fork
Migrations are rarely a clean cut. Half the team moves, the other half finishes something on the old tool, and for a few weeks both are in use — which is normally when the two setups begin to disagree, because each accumulates its own version of what was settled. Here both read and write one set, so a rule settled by whoever moved first is already in force for whoever has not, and the migration stops producing a second truth for someone to merge later.
What it looks like in practice
A team decides to move from one assistant to another. The plan covers repositories, permissions and billing, and nobody lists context, because "the prompts are in the repo". Six weeks in, the new assistant proposes a pattern the team ruled out in March. The reason it was ruled out came up in a session, was summarised into the old tool's memory, and was never written anywhere else. Nobody noticed it was gone, because a missing decision does not raise an error. It produces a reasonable-looking suggestion. The repositories migrated fine; what did not migrate was the only thing the plan never listed, because it did not look like an asset.
Questions people ask about this
- Which tools can read the same memory?
- Any tool that speaks the same connection standard — Claude, ChatGPT, Codex and others connect to Arroway as a server, and there is no per-tool copy of the material. So "supported" means "can connect". How the read is triggered varies by tool, and where a tool cannot be made to read at the start of the work on its own, the short human review is what covers the gap.
- Can we take our material out again?
- Yes. Export ships with the product rather than being a support request. The point of keeping decisions outside the tools is that they are yours, and that has to include leaving — a memory layer you cannot walk away from would reproduce exactly the lock-in that makes this migration expensive.
- What about the memory our old assistant built up on its own?
- That lives inside the account and scope its vendor defines, and this is not a way to extract it. What you can do instead is deliberate: go through what actually mattered and write down the parts that were team decisions, with a person approving each one. It is shorter than it sounds, because the test is narrow — a decision someone would be wrong to contradict, not everything that ever came up.
Where this is verifiable
Product documentation on this site (How it works, Install), the open connection standard the server implements, and the export tooling that ships with it. Everything described here is behaviour the tools apply today, not roadmap.
https://www.arroway.app/en/answers/switching-ai-tools-without-losing-team-decisions