What Capsa is, why it matters, and exactly how it works.

You already did, by accident. People tell AI what they tell no one: the 2am health worry, the resignation letter never sent, the business idea, the draft of a hard conversation. Nobody sat down to keep a diary, but after a year or two of talking to AI, one exists. It is the most honest and complete record of how you think that has ever existed for anyone. Not chat logs. The story of your mind during the years everything changed.

In 2005 nobody thought phone photos mattered. Today the camera roll is the first thing people would save from a fire. Your AI conversations are the camera roll of your mind. The difference is that the vendor holding it deletes, rewrites, and retains on its own schedule.

You have never read it. What do you really ask about? When do you think best? Do you push back on the AI or accept whatever it says? People pay real money for dashboards about their sleep and their runs, built from data they were producing anyway. Your archive is the only self knowledge instrument that requires doing nothing new. Capsa reads it and shows you yourself: what you use AI for, whether you get what you came for, who is influencing whom. Always against your own baseline, never against other users.

The models are becoming commodities. Everyone has access to the same ones, so what differentiates people is how they think with them. The record of everything you have said and decided is the richest input to a personal AI that actually knows you, and it is the raw material for understanding and improving how you reason. Your usage is spread across many tools, and no single tool holds the whole of it. You are currently generating the most valuable dataset of your life, about the only subject no one else can ever own. Capsa exists so the complete record sits with you.

Because an archive cannot be backfilled. Claude Code deletes its records after about 30 days by default. Vendor memory profiles get rewritten and are not exportable. The last independent personal AI archive was bought and shut down within two weeks. Every unarchived day is a page of the diary gone, and unlike a backup, you cannot write yesterday’s entry tomorrow.

Extend and improve human thought and reasoning. That is the destination, and the way there is concrete: collect your AI conversations from every tool, archive them in their original form under your control, and harvest the reasoning patterns inside them, the direction you gave, what you accepted, what you rejected, what you corrected. Today that gives you a complete record, readings of how you work, and evidence of your contribution. As the record grows it becomes training data you own, the seed of an AI that judges like you. Parts work today. The rest arrives as your record grows.

Two layers. First, your originals: every file, byte for byte, exactly as your tools wrote it, including a dropped ChatGPT export ZIP kept whole. This is the archive itself, and nothing is interpreted or thrown away. Second, a readable view derived from those originals: conversations you can browse, and statistics. The readable layer can always be recomputed; the originals are never touched.

Capture and understanding are separate on purpose. Your Copilot and Cline records are banked in full today even though their readers are still being built. When a reader lands, it runs over what was already banked and those conversations appear retroactively. Bank now, understand later. The one thing that can never be recovered is data that was not captured.

Nothing leaves without a deliberate act by you: running capsa sync, or dropping an export file onto the page. The local folder the CLI creates on your machine stays yours and keeps working even if you never open an account.

Yes, if it is in your cloud account: an AI model reads each conversation once and writes structured notes: what it was about, what you came to do, whether you got it, how you engaged. These notes power your Insights and the Ask feature. The analysis runs under our API key, is never used to train anything, and every reading keeps a verbatim quote from your record so you can check it. Counts and hours are computed exactly, not by AI.

Never. Capsa only ever shows you to yourself. Every comparison on every page is against your own history.

You, behind your sign in. We do not browse user archives. The content is stored on our servers and is not yet end to end encrypted; an option where you hold the encryption key is on the roadmap, because sovereignty should be engineered rather than promised. If that is a requirement for you today, keep sensitive sources local and sync only what you are comfortable holding in the cloud.

Two buttons in Settings, never buried. Export all returns everything as one file: your originals byte for byte plus your conversations as standard JSON. Delete everything removes your account and every file immediately and permanently, with no grace period. Your exact bytes back, or truly gone. Your choice.

Fully today: ChatGPT, through its official data export dropped onto the page, and Claude Code, through the capsa command line tool, which reads the records Claude Code writes on your machine. GitHub Copilot, Cline, and Cursor records are banked in full by the same tool as originals; their readers are still being built, so they are stored but not yet browsable. A browser extension that banks ChatGPT conversations as you have them is built and coming to the Chrome Web Store. Support for claude.ai and Gemini is being built: you can take a claude.ai export today and bank it the day support lands. The guides on the Sources page show what each tool keeps and how to get your copy out.

By fingerprint. Every file gets a cryptographic hash; the server answers which fingerprints it lacks and only those files travel. The first sync sends your whole record and can take a while. Every sync after sends only what changed and takes seconds. The server verifies every fingerprint on arrival, so a corrupted transfer can never silently enter your archive. Interrupted syncs resume where they stopped.

Every conversation becomes a session card built from the facts of your record: what you asked, how the work moved, what you accepted, rejected, or corrected, and what was produced. There is no score and no grade, just the facts of the session. The Echo is the daily read over the record: what is on your mind, how your thinking is moving. Every reading keeps a verbatim quote from your record so you can check it against your own words.

Chat with your own archive: what did I decide about X, when did I first look into Y, show me where I changed my mind. Answers come only from your record, with citations to the exact conversations, and it says plainly when your record does not contain the answer. Ask is part of the premium plan.

The local tool is free and stays free. During the pilot the cloud archive is free too. Paid plans arrive with the public launch and the premium tier carries Ask. Pricing will be published plainly before anyone is charged.

Export all works today and always will: one click returns your complete originals. Your local vault never depended on us in the first place. An archive business that held your data hostage would contradict the one thing it exists to do.