FAQ
Can I see it in action without installing anything?
Yes. The Recall and User Card docs have questions you can click right on the page — Previously answers them from a real two-and-a-half-year memory. No install, no key.
Where does my memory live?
It depends on which way you run it. Local (npm): at ~/Documents/Previously, a directory of plain Markdown files plus a local git repository. Deployed to Vercel: in your own private GitHub repo, as the very same files. Neither form has a database, a vector store, or a proprietary format — you can open it in any editor. What it actually looks like on disk is in Your Memory Is a Folder.
Who can see my memory?
By default, only you and your model provider. In the local (npm) form your conversations never leave the machine; the single external path is the model itself — bridge mode rides your local agent CLI subscription, BYOK mode rides the provider key you configured, and that exposure is inherent to using any hosted model. When deployed to Vercel, memory sits in your private GitHub repo, and who can see it comes down to repo permissions and how well you guard the deployment URL (the cloud has no account system — set ACCESS_SECRET if needed). There is no Previously server and no platform database — we hold nothing.
Does it forget the beginning of a long conversation?
No, because there is no single conversation that grows without bound. Context is reassembled on every turn: the system prompt is built in layers and frozen byte-for-byte within a slice (which is why precise time comes from the currentTime tool on demand instead of being injected), and the past in the timeline is pointers, not payloads — when something looks relevant, the agent opens that slice and reads it on the spot. The archive can grow across months; what goes into any single request stays window-sized.
What does it cost to run?
Previously itself charges nothing: no subscription, no per-seat fee, no markup. What you pay for is the brain — in the local (npm) form, bridge mode reuses your existing Claude/Codex/Kimi subscription at zero marginal cost; BYOK mode is your own API key and your own bill, with an architecture designed to be cheap (inexpensive steps do the housekeeping, the strong model is reserved for the conversation). The full comparison is in Two Engines. Deploying to Vercel adds whatever your Vercel plan costs; GitHub repo storage is effectively free, and the brain is likewise your own provider key.
Can I move my memory elsewhere?
Any time, because it was never "inside Previously" to begin with. Memory is a git repository: clone it, push it, pack it up, or just point another tool at the directory. Moving away needs no export feature — there is nothing of yours being held.
What about my existing Claude/Codex/Kimi history?
It can be transcribed in. The client's scribe incrementally transcribes those agents' session logs into Previously time slices — transcribe, don't hijack; the original tools keep working exactly as they did. The mechanism is in The Scribe.
Is it finished?
Not yet — honestly not yet. Previously is in active early development: expect rough edges and breaking changes. But the core discipline — memory as files, an auditable write path, your data belonging to you — has been an architectural constraint since day one, not a roadmap item. It is a personal project, not a SaaS.
Related
- Your Memory Is a Folder — what memory actually looks like on disk
- Two Engines — the full bridge vs. BYOK comparison
- The Scribe — how existing agent history becomes slices