01
You point your agent at it
One command, then a browser sign-in. No server to run. No extension to maintain.
Project memory for AI-assisted work
A board that keeps every project current, so every connected AI session starts knowing where you left off and what you already decided. You stop being the only thing holding it together.
14 days free, no card. Your trial starts when you connect your first agent.
Or look at a sample board first — no account needed
Live hub in action.
Tested end to end
Should work, not yet tested
The Hub also ships MCP prompts, so it turns up as slash commands inside your own editor: catch-up, pick-up, wrap-up, and needs-me.
You glance at the board. The AI keeps it accurate.
01
One command, then a browser sign-in. No server to run. No extension to maintain.
02
Every session opens with the project's next action, where the last one stopped, and the decisions already settled. Nothing to re-explain, nothing re-litigated.
03
When your AI hits a call only you can make, it pins the question to the board and carries on with what does not depend on it. You answer once, whenever you get to it.
04
When the work ends, the board updates on its own. Not because someone remembered to, and not only in the tool you happened to be using.
Every tool you already own needs you to maintain it. This one maintains itself, because the sessions doing the work are the ones writing it down. That is the entire difference, and it is why the board is still accurate three weeks later when you come back.
A place to keep notes
Astryke
You keep it up to date
The sessions doing the work keep it up to date
Notes drift out of date the day you stop writing them
Current state is written back at the end of every session
Context lives inside one AI tool
Decisions follow you across tools
Questions for you get lost in a chat log
Decisions that need a human wait on the board until you answer
The Hub holds what you are building and what you decided, which is unusually personal for software. Here is exactly what that means.
Connected agents send structured project updates, not your project files. The Hub has no access to your machine and cannot go looking for anything. Agents do write those summaries and questions themselves, from their working context, so it is worth knowing what your chosen AI tool sends.
Session summaries, decisions, questions and answers, pinned thoughts and replies, and session records are encrypted. Project names, the one-line next action, and working-directory paths are not, because the Hub matches projects by name on every agent call and cannot do that against ciphertext.
Every connection uses its own token, with created and last-used history visible in the app. Revoke one and it stops reading or writing immediately.
Your projects and notes are not training data, are not sold, and are not handed to anyone else. There is no advertising and no third-party analytics script inside the app.
No tiers, no seats, no feature gates. As many projects and agents as you want.
Founding price
$12 / month
or $120 a year, which is two months free.
Locked for as long as you stay subscribed. The price goes to $15 a month after the first 100 accounts.
Start free14 days free, no card. Your trial starts when you connect your first agent, not when you sign up.
When the trial ends, your board stays readable. Forever, by you and by your agents. What stops is writing new notes. Nothing is deleted and nothing is held hostage.
You can take everything with you. Export the whole workspace as JSON or Markdown at any time, whether or not you are paying. Deleting your account removes everything in it, and cancels any live subscription first.
Switching between them costs a re-briefing every time. The board holds each one's state so the cost is a glance instead of a conversation.
What you settled in one does not exist in the next. Decisions live on the board rather than inside whichever assistant happened to hear them.
Being the only thing holding a project together works right up until it does not. This is the part that can be handed off without handing off control.
It never scans or uploads your repository, and it has no access to your machine, so it cannot go and fetch anything. What it holds is what your agent sends: project status, next actions, decisions, questions, and the working directory it was in. Those are written by the agent in its own words, so a summary or a question could quote a line of your work if the agent thought it mattered. Worth knowing when you pick which tool to connect.
Claude Code, Codex and VS Code with Copilot are tested end to end. Cursor, Devin Desktop (formerly Windsurf), and other MCP clients should work over the same protocol but have not been verified yet, and this page will say so until they are. Setup is a config entry and a browser sign-in.
What it worked on, where it stopped, what it decided, and anything it needs you to settle. It also writes a session record so there is a history of what happened. You can read, edit, or clear any of it yourself.
Yes, at any time, from the Connect page in the app. Every connection has its own token with created and last-used history. Revoking one stops it reading or writing immediately, and does not affect your other connections.
The board stays fully readable, forever, by you and by your agents. What stops is writing new notes. Nothing is deleted, nothing is locked away, and nothing is held hostage to get you to subscribe.
Both, yourself, at any time. Export the whole workspace as JSON or Markdown whether or not you are paying. Deleting your account removes everything in it and cancels any live subscription with Stripe first.
Make a workspace, point your next AI session at it, and the board starts keeping itself current. 14 days free, no card. If it does not earn its place, walk away and nothing was charged.
Create your workspace