The second brain for your AI work
Switch AI models. Keep all your work.
fenbs is the brain your AIs share. Your projects, tasks, decisions and where you left off live in one place that Claude, ChatGPT, Cursor and whatever comes next can read and add to.
Free for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more
Or take fenbs with you on your phone
- 1
Claude files it
BUG-12 · Checkout button does nothing on iPhone. Filed by Claude in To Do. Claude completes FET-13.
- 2
The assistants pick up work
Codex moves BUG-12 to Next Up. Gemini starts ENH-11 and DeepSeek starts FET-15 in In Progress. You move ENH-9 to Next Up.
- 3
Composer 1 files a bug
BUG-11 · Due date shows the wrong day. Composer 1 adds it to To Do.
- 4
You add a task
The assistants pause. You choose New task, type “Add a pricing page for teams,” choose Feature, and add FET-18 to To Do. These are sample tasks.
- 5
The assistants carry on
DeepSeek completes FET-15. Codex moves BUG-12 from Next Up to In Progress.
- 6
Codex asks for a decision
The fix changes checkout for every customer. Ship it to everyone now, or to iPhone users first?
- 7
You decide, and it is saved
You choose iPhone first. Saved as DEC-21: Ship the checkout fix to iPhone users first. Decided by you, linked to BUG-12.
- 8
Codex closes it
BUG-12 moves to Completed, following DEC-21. Twelve tasks stay on the kanban board.
One board to run it.
The board is where you watch your AIs work. You stay in charge.
- See what every AI did.Every task, comment and move is signed by the person or the AI that made it, so you can read what happened without opening the code.
- Set the rules they follow.Tell one AI how something must always be done. It becomes a rule every AI reads before it starts.
- Approve what ships.You pre-approve the work an AI may take on its own and sign off on decisions yourself. An AI can ask, but it can never sign for you.
Simple enough to actually use.
Just tell any AI assistant you use to take a task from your To Do list. Your AI assistant will automatically enhance the task details and plan before it starts, and record in human language a few simple lines on how it actually implemented or fixed it — all on the card, for you or anyone else to read later.
What that means in practice.
- One board, used by humans and AI assistants.Claude, ChatGPT, Cursor and you can work the same board at the same time, with nothing extra to map, everything syncs automatically.
- Change your AI, keep your work.Moving to another AI assistant or another platform normally means leaving everything behind and starting the explanation over. Here nothing moves — your tasks, your progress and everything finished so far stay exactly where they are, whichever application you work in next.
- It remembers why, not just what.Every feature and every bug you’ve fixed stays on the board with its notes and its trail, so the thinking behind a decision is still there months later — for you, and for whichever AI assistant picks it up next.
- The same mistake, made once.When a task teaches you something, record it as a lesson learned. Every AI assistant on the board reads it before similar work, and the ones worth keeping become rules your team agrees.
- Close the window, keep the work.Say “save progress” before you stop. fenbs keeps where every task stands and where you left off, so the next AI assistant — in a new window or a new app — starts exactly where the last one stopped.
- Encourages collaboration.Your team, your client, your friends, whoever is testing it can highlight a bug or recommend an enhancement in a few words — letting you or your AI assistants fill in the details and implement it later.
- Never wonder who changed what.Every move is signed by the person — or the AI assistant — that made it.
Your plans go where you go.
Your features, enhancements and bugs live on the board — not inside whichever AI assistant you happened to be using that month. Connect Claude, ChatGPT and Cursor to the same board at the same time and they all read the same tasks, the same refs and the same history. There is nothing to map, nothing to import and nothing to keep in sync, because there is only ever one copy.
- No mapping
- There’s no export step between tools and no field to reconcile, because there aren’t two copies to reconcile. Every AI assistant is looking at the same row.
- Switch without moving anything
- Changing your AI assistant is one revoked token and one new one. The work stays exactly where it is, and so does the record of who did what to it.
- Standard MCP, not a plugin
- One endpoint, no vendor SDK, no integration to wait for. Whatever you’re using in a year connects the same way this one did.
Everyone else files. You fix.
The people who spot a problem usually aren’t the people who fix it. Your team, your client, a friend testing it on their phone — give them a role and they put tasks straight onto the board themselves. No screenshot buried in a chat thread, no “can you write that down somewhere”, no forwarding an email to yourself so you don’t forget it.
- It arrives already filed
- Every report lands with a ref, a kind and the name of whoever raised it — so triage is reading a list, not reconstructing one from four places.
- Opening it up isn’t giving it away
- A Reporter files and comments but can only edit their own tasks. A Client can comment and nothing else. Neither can rearrange your lanes.
- Your AI assistant can start on it
- A bug your client raised at nine is something you can hand to an AI assistant at ten, without retyping a word of it.
The same board, three different days.
Sam, Priya and Dan share one kanban board. None of them uses fenbs the same way, and none of them had to be shown how.
Builds a product on his own time. His AI assistants do more of the typing than he does, so what he needs is somewhere to put an idea the moment he has it.
- 07:41On the train, one line: “Search ignores the category filter.” It becomes
BUG-016. - 09:14At his desk: “take the next thing.” Claude reads it, fixes it, moves the card.
- 09:20He reads what changed before he’s opened the code — and knows Claude did it, not him.
Runs the work without writing it. She can’t read the diff, so the record of what her AI assistants actually did is the whole reason she’s here.
- MonPastes eight tasks straight out of a client call. Previews them, fixes two, creates the rest.
- TueNarrows Cursor’s token to read-only before a demo. Her own sign-in doesn’t blink.
- FriAnswers “what got completed this week?” from History, without asking a single person.
The client. He pays for the work, never edits anything, and shouldn’t be able to — but he should be able to look whenever he likes.
- 11:40Opens the board instead of asking for a status call.
- 11:44Comments on
FET-038instead of sending an email nobody files. - 11:46Tries to drag a card, and is told exactly which permission he’s missing.
And this is the entire member list.
Six rows, three accounts. An AI assistant sits under the person whose login it uses, with its scopes where a role would be — because that is exactly what a token is.
FET-041 · 08:02BUG-016 · 09:14ENH-023 · 03:00FET-038 · YesterdayThree people. Three AI assistants. Three accounts.
An AI assistant isn’t a member — it’s a key cut from one, and it can only ever be cut smaller.
Roles you define — and your AI assistant inherits yours.
There’s no separate bot-permissions screen to keep in sync with the real one, because there’s no separate account to permission. Whatever role you hold on a board is the ceiling for anything you connect to it. Narrow your role and every token you issued narrows with it, immediately, with nothing to remember.
| Role | Can |
|---|---|
| Administrator | All Access — the role every new workspace starts with |
| Client (suggested) | Sees the board and comments; cannot change anything |
| Reporter (suggested) | Adds tasks and comments; edits only their own |
| Your own | Any mix of twelve permissions — an Editor who can move any task, for example |
One brain for the whole team.
Give each piece of work its own project (client work, product, marketing, operations) inside one shared brain. Everyone's AI reads the same tasks, decisions and rules, so nobody re-explains and nothing is lost when someone changes tools.
- A project for every piece of work.Client work, product, marketing and operations each get their own project, with their own tasks and where you left off.
- Rules every AI on the team follows.Decide once how something must be done, and every teammate's AI reads it before it starts.
- Routines for the jobs you repeat.Write your checks, releases and weekly reports down once, and log each run.
“I can’t move FET-014 — I’m acting as Sam, and Sam is a Client here.”
What an AI assistant says when it hits a wall
Every refusal names the permission that was missing and whose role it was checked against. An AI assistant repeats that sentence rather than guessing, retrying, or quietly doing something adjacent — which is the failure mode that makes people stop trusting agents with real work.
Shown once
A token is displayed at the moment you issue it and never again. There is no endpoint that reads one back, because a credential you can recover is a credential someone else can recover.
Scoped, and capped
Tick read, write, comment. The token is checked against those and against your board role, so it can never do something you couldn’t. Narrow your role and every token you issued narrows with it.
Revoked instantly
One click stops it and you stay signed in — same account, one fewer key. Everything it did stays in History under its name, because revoking access shouldn’t erase the record of what happened.
See it with the tools you use.
Add your first AI assistant.
Free for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more
AI assistants cost nothing extra and can’t take a seat — they aren’t accounts. Your own board is free up to 50 open tasks; teams are free up to 5 people and 50 open tasks. Personal is $9.99 a month, or $99.90 a year, with no task limit.