On your device
Reading your board
The board summary is not a trial and not a teaser. It runs locally, makes no network call, spends no credits and works with the connection off.
- No network call
- No credits
- Same board, same summary, every time
Launch 20% off every paid plan for the first year.
See the plansAssistant
Summarising a board costs nothing and never leaves your machine. Everything that does call a model tells you what it costs before it runs, in credits, per action. And an administrator can switch the cloud half off for the whole workspace.
Two halves
Most tools call a model for everything and bill you for the privilege. Here the reading is local and free, and only the writing costs anything.
On your device
The board summary is not a trial and not a teaser. It runs locally, makes no network call, spends no credits and works with the connection off.
In the cloud, on request
Generating a table, a plan or a set of tasks needs a model. Each of those is a named action with a fixed price, a size limit and a result you review before anything lands on the canvas.
In practice
Not "productivity". Specific work that used to be retyping.
2 credits
Forty stickies, three frames and a decision nobody wrote down. Task extraction reads the canvas and proposes real Flowtask tasks, each keeping a reference back to the object it came from.
1 credit
Amount, date, vendor and tax come out of the photo, so the expense record starts filled in. You still check it, but you check a form instead of typing one.
Free
A board nobody has opened in six weeks, summarised on your own machine in the time it takes to read this sentence. No account decision, no balance, no connection.
Capabilities
Twelve actions, not one chat box. Each is a defined job with a fixed price and a size limit, so nothing can quietly turn into an expensive request.
Reads the stickies, frames and notes on a canvas and proposes real Flowtask tasks from them, so a workshop does not have to be retyped into a planner.
Takes a goal and the board's own content and proposes a plan of concrete board actions. You review the proposal, and nothing is applied to the canvas without you.
Tables, mind maps, flowcharts and presentation outlines, produced as real editable board objects rather than a picture of a diagram.
Place a generated image straight onto the canvas when you need a placeholder or a reference frame in the middle of a session.
Describe the project in a sentence and get a first Flowtask breakdown to edit. It is a starting point for the plan rather than a substitute for making one.
Pulls amount, date, vendor and tax out of an expense receipt so the expense record starts filled in instead of blank.
Tighten a card title, summarise a short note, or sort a line of text into the right task type. Small jobs, priced like small jobs.
Requests are built from the board content that is relevant to the question, and answers point back at the objects they came from, so a claim can be checked against its source.
Cost per action
In credits. The price is fixed per action and does not vary with the length of the answer, so a bill cannot surprise you after the fact.
| Action | Credits |
|---|---|
| Board summary (on your device) | Free |
| Rewriting that summary in the cloud | 0.5 |
| Table generation | 1 |
| Expense receipt extraction | 1 |
| Short text rewrite | 1 |
| Task classification | 1 |
| Short text summary | 1 |
| Mind map generation | 2 |
| Project plan generation | 2 |
| Task extraction from a board | 2 |
| Flowchart generation | 3 |
| Presentation generation | 3 |
| Image generation | 3 |
| Copilot planning | 3 |
Credits are reserved before the request is sent and settled after it returns. A request that fails does not consume the reservation.
Governance
This is not a preference each person is asked to remember. It is workspace configuration, enforced on the server.
One switch for the whole organisation. The free on-device summary keeps working, and nothing else calls a model at all.
An allowlist per workspace. Permit receipt extraction and short rewrites, refuse image generation, or whatever else fits your policy.
A monthly token quota for the workspace and a per-person monthly credit ceiling, so one enthusiastic week cannot drain the budget.
AI can be disabled for named accounts while the rest of the workspace keeps it, which is what you want for contractors, shared logins or a role that should not send content out.
Cloud actions go to a single model family on its first-party endpoint, with provider fallbacks switched off. A request is never quietly rerouted to a different vendor because the usual one was busy. It fails instead, and a build-time guard rejects any configuration that points somewhere else.
Every cloud request passes the same stack before it reaches a model: authentication, the workspace AI policy, a payload check that the content is text of the expected shape and size, and a credit reservation.
No. There is no open-ended chat window. Every action is a defined job with a fixed price, a size limit and a reviewable result. That is a deliberate trade. There is less to play with, and nothing that runs up a bill you did not choose.
No. Only the board summary runs locally. Everything in the price table above needs a connection and a paid plan. Your boards themselves stay editable offline either way.
No. It selects and ranks sentences that are already on your board. That is what keeps it free, deterministic and offline. The optional half-credit refine is the step that rewrites.
The on-device summary is available to everyone. The cloud actions require a Pro plan and credits to spend.
Summarise a board without an account decision, a credit balance or a connection. Add the rest when it earns its place.