Ten tabs, no state
Every tool keeps its own history. None of them knows where the others stand — and you are the integration layer in between.
For macOS · Founding Beta One mission.
One mission.
One AI team.
One reviewed result.
For people and teams coordinating multi-step AI work.
You tell Undernap the outcome you want. Undernap turns it into a Mission: a plan you approve, tasks for specialists — executed over the AI accounts you already have — an independent review, and, at the end, a result with evidence.
Founding Beta for macOS (Apple Silicon, macOS 12+). Nothing is sold yet and there is no public download — access goes through the Founder Beta.
1mission instead of ten tabs
5specialist roles
1reviewed result
The problem
AI work spreads out. The overview does not.
Anyone working with AI seriously ends up with several tools: a chat model here, a coding agent there, a second account for capacity. That works — until the job outgrows a single window.
Copy-paste as a job
Shuffling results by hand, explaining the context twice per tool. In the end nobody can say which version counts.
“Done” is a claim
An answer sounds good. Whether it matches the job, nobody checked — and that lands back on you.
Capacity evaporates
Hit the limit of one subscription and everything stalls, while a second account sits there with capacity to spare.
The difference
Not another chat window. A workspace that coordinates.
Undernap does not replace your AI tools — it puts them in order. Five things change, concretely.
Today: tab to tab
You break the work down yourself, carry the context from window to window and keep the state in your head.
With Undernap: one Mission
One request with a plan, tasks, ownership and a defined end. The state lives in the product, not in your memory.
Today: model roulette
Deciding afresh at every step which model is probably best at this part.
With Undernap: specialists
Roles with a quality bar — research, implementation, review. Which connection runs a role is settled by the plan.
Today: loose accounts
Several subscriptions that know nothing about each other. One is at its limit, the other is idle.
With Undernap: a Fleet
Your connected AI accounts as one pool of work capacity: capacity becomes visible, roles get staffed, limits get respected.
Today: blind trust
You read the answer and hope it holds. Checking it means reading everything yourself.
With Undernap: review
A separate instance checks against the goal before you see the result — never the one that did the work.
Today: scattered answers
Five windows, five partial truths, and you do the merge.
With Undernap: one Result
One version that counts — with the evidence behind it and your acceptance at the end.
How a Mission runs
From idea to result in five steps
You stay the deciding authority at three points — at the start, before anything runs, and at the end.
- Give the request
The goal in your words — plus limits: read-only, nothing deleted, budget.
You decide - Approve the plan
Undernap proposes the plan, the roles and the limits. Nothing runs before that.
You decide - Let it execute
Specialists work over your AI accounts — in bounded sessions that leave evidence.
- Let it be checked
Verification and an independent review decide whether “done” is actually done.
- Receive the result
One result with its evidence. Open points wait visibly for you.
You decide
The interface
Four screens, one flow — from request to verified result.
A design preview of Undernap’s dark interface: layout, terms and flow follow the real screens of the running Founding Beta — whose interface is still light today.
The previews are built on the real Founding Beta screens and show nothing the product cannot do today. Real, unretouched captures — carrying the development name the app still shows — are further down beside the Result chain and on the Result & evidence page.
Why you can trust the result
“Done” is a verification result. Not a claim.
The strongest part of Undernap is not that it distributes work — it is that you can trace what happened to it. Every Mission runs through the same chain:
- Execution Receipts
Specialists work through the tasks in bounded sessions. Every session leaves structured evidence.
- Verification Checks
Results are held against the request: does what is claimed match what is evidenced?
- Independent review Review
A separate instance decides — never the one that did the work. A reviewer from the same provider family as the executor is refused.
- Canonical result Result
Only when the review confirms against the goal does the Result become binding — with the evidence behind it.
- Your acknowledgement Owner
You see the result, the evidence and the open points — and your acceptance closes the delivery.
Every statement carries its kind: observed, derived, inferred or unknown. What is not evidenced is said — not glossed over.
What you would use it for
Concrete jobs, not “productivity”
Three of the roles that exist in Undernap today — and what comes out of them.
Check a repository before a release
“Analyse this repository read-only and name the three biggest product risks.” Out comes a risk report with file-level evidence — checked by an instance that is not its author.
A website audit with sorted findings
Technical, content and discoverability findings — each statement marked observed, derived or inferred. No metric without a provenance.
Market and competitor research
A briefing with what is evidenced — clearly separated from what was merely concluded. No invention presented as fact.
Several AI resources, one order
Your Fleet: several accounts, several strengths — instead of chaos.
Undernap does not sell AI usage. You connect the accounts you already have; together they are your Fleet. The value of several accounts is not status, it is division of labour:
Proven in operation through the product: Claude, Codex, Grok, MiniMax, Z.ai GLM and OpenRouter (free pool). Further providers are qualified one by one before they count as usable.
Roles meet strengths
Research, implementation and review are different jobs. Different accounts and models may suit them differently well — the role still stays the role.
Capacity is a fact
Two accounts with the same provider are two parallel workstations. When one hits its limit, Undernap waits or moves within your rules — limits are respected, never circumvented.
Independence is enforced
For the independent review: an instance from the same provider family as the executor is refused. Real checking needs real distance — that is built in, not promised.
Starting with one is fine
Just as valid. The Fleet grows when you grow — Undernap does not require a second subscription to be useful.
Your control
A workspace that asks before it acts.
Automation without control is just a faster route to the wrong result. Four boundaries are therefore settled in the product, not in a promise.
You approve
The plan, the judgement calls along the way and the acceptance at the end are yours. Nothing closes itself.
No surprise spend
Paid execution needs clear limits and your approval. There is no quiet consumption in the background.
No silent widening
No provider is added that you did not connect. Your Fleet is exactly what you set up.
Local first
Undernap runs on your Mac. Project knowledge, missions and results live in your local data directory — not in our cloud.
Price & access
The final price is not settled. The rules around it are.
Undernap is a Founding Beta — it cannot be bought yet. What is already settled is the economics of the product:
See the pricing page Founding Beta · macOS
- One licence covers one installation — verified offline, with no licence server.
- Your AI subscriptions stay yours: Undernap sells no AI usage and earns nothing from what your accounts consume.
- Founding Beta licences allow permanent use of the product. What is time-boxed is beta access and the right to updates.
- Deliberately still open: the final price, subscription versus one-off, seats for teams. Those are decisions to make before release.
Be early
The Founder Beta opens in small steps. Ask for a seat, confirm your email — and you will be among the first to hear when access becomes possible. No advertising, unsubscribe any time.
Founding Beta for macOS (Apple Silicon, macOS 12+). Nothing is sold yet and there is no public download — access goes through the Founder Beta.