Data storage and privacy
Causal holds what your agents learned: the decisions they made, the outcomes that followed, and the reasons behind both. This page covers where that lives and what you can do with it, on any self-serve plan.
Where is my agent memory stored?
On Causal's servers, encrypted at rest and in transit. One graph in one place, which is what lets a memory written in Claude Code come back in Cursor the next morning.
Your machine keeps a small working cache — recent writes on their way up, recent results already served — so an agent never waits on the network mid-task. The cache is a convenience. The graph is the record.
What does Causal actually store?
What your agent hands it. Causal does not read your filesystem, index your repository, or watch your screen. Capture is a call the agent makes when something is worth keeping: a decision, the outcome that followed, context that will still matter next week.
So a memory carries whatever the agent judged salient — the shape of a fix, the error that came back, the reason a path was set aside. The salient part of the work rather than the transcript of it, which is also why a briefing arrives in a few lines instead of an hour of conversation. Recall and capture covers what gets written and when.
Is my data sold, or used to train models?
No. Your memory is yours: never sold, never shared with another customer, never used to train models.
Turning raw activity into experience runs through a model, so captured content passes through Causal's extraction pipeline and the model provider behind it. Neither trains on it.
How long is it kept?
As long as you want it. There is no expiry on a memory and no clock on an account.
The free grant never refreshes, and it also never takes anything away. When it is spent, writes pause and everything already stored stays exactly where it is — search, lineage, and verification keep working on it, permanently. A free account left idle for ninety days reads more slowly until the next write; nothing moves and nothing becomes unreachable.
If a subscription lapses, the graph goes read-only. It is not deleted.
How do I delete something?
Two ways, and the difference is worth a moment.
- Retire it from recall. The memory stops being served and stops reaching decisions. It stays in the history, so the trail can still explain what your agents did while it was live.
- Erase it. The content is destroyed and unrecoverable, backups included. What survives is the outline of a record that was deliberately erased, which is what keeps the rest of the graph verifiable.
Retiring is the default, and usually the one you want. The window in which agents acted on a lesson that has since been corrected is often the part of the answer you need later. AI agent governance and audit trail covers what that trail can rebuild.
The whole account can be deleted whenever you want.
Can I take my graph with me?
On any paid plan, yes. The graph exports in an open format, and the standalone causal-verify checks that export offline — against our word rather than with it. Verification itself is free on every plan, including the free one, because the claim the product rests on should not sit behind a card.
Being easy to leave is why the state is worth handing over in the first place.
For larger teams
Data residency, your own cloud or VPC, bring-your-own keys, retention policy, and role-based access are enterprise controls, set per deployment rather than by a switch on a plan. AI agent governance and audit trail covers them.
Everything on this page holds on every plan below that, free included.
What your agent is served
Storage is one half of the question. What actually reaches an agent when it recalls — and how personal work stays separate from a team's — is on Selective recall and memory scope.