Run it in the cloud or on your own machine. Agents handle the busywork in a safe sandbox — you approve every change before it counts.
The whole workbench runs on your computer. Click around — start a thread, watch it run, review the packet.
Every agent gets its own private, sealed copy of your files. It can read, draft, and sort — but it can’t change your real documents. It plays in the copy, never the original.
When it’s done, you see a review packet: what changed, in plain words, with the source behind every claim. No jargon, nothing to decode.
Only when you say yes does anything touch your real files. Apply all of it, just the parts you want, or none. Every step can be undone.
There’s no technical setup and nothing for you to manage. The agents work offline on a sealed copy, every change is logged and reversible, your originals are never touched — and nothing leaves your computer unless you choose to send it.
That’s the floor every install gets. Teams in regulated or untrusted environments can lock the network down too.A pack is a set of ready-made playbooks for one kind of work — the reports, reviews, and busywork you already do. Open one and you’re running, no setup, no prompt-writing.
for project managers, delivery leads & ops
Status reports, ticket triage, standup digests, and risk logs — drafted from your tools, reviewed by you.
Explore the packfor solo & small-firm lawyers and paralegals
Drafting, deadline tracking, e-file archiving, and fact extraction — cited and confidential, on your machine.
Explore the packfor pharmacies, compounding & QA teams
Batch-record QA, SOP conformance, and inventory reconciliation — gaps flagged for a licensed sign-off.
Explore the packplaybooks.cool keeps the substrate that makes agents safe — isolation, audit trails, review-before-apply — and hides everything that makes them feel like engineering.
Every proposed change — from an agent, a playbook, or a teammate — arrives as a review packet: a plain-language summary, the grouped changes, the evidence behind each claim, and the approvals it needs. Approve, apply some, request changes, or discard. Nothing touches your files until you say so.
Documents, context, playbook runs, and everything the system learns about your work stay on your computer. No cloud upstreaming by default, no training on your data. Import, run, and export all work with networking switched off. The agents run offline and never need to phone home.
Drop in a transcript, a document, or a finished project and the workbench distills it into a versioned, reusable playbook — not a saved prompt. Every run pins its version, so last month’s work stays reproducible even after the playbook evolves.
Deadlines, amounts, names, totals — anything your policy marks critical requires individual human sign-off, with every claim anchored to a source document. No bulk-approve. No made-up fact gets past the gate.
Hand a packet to a colleague as a file, or pair two computers over your own network. Presence, pending work, and overlapping changes show up as people and proposals — never as a confusing technical mess for someone to untangle.
Route each run to Claude, GPT, or a local model, and dial reasoning effort per task — Plan, Fast, or Ultra. Every run records the model, the playbook version, the approvals, and the evidence. Your team’s quality floor is never silently undercut.
Quotes from practitioners across project, legal, and pharmacy teams in our private beta, attributed by role.
It writes the status report I used to spend Friday afternoon on — and I can see which standup every line came from.
It drafts the brief, but shows me exactly which document every fact came from. Nothing gets filed that I didn’t check myself.
It catches the missing initials and the lot that’s about to expire — then it waits for me to sign.
There’s nothing technical to set up or wrestle with. It just works the way I already think about my work.
My team passes work back and forth over our own network. The data never leaves the office.
It learned how we run things and turned it into a playbook the whole team can use.
No. There’s nothing to install and no setup — you work with workspaces, threads, save points, and review packets, plain words that match the work you already do.
Everything stays on your machine by default. The agents work offline on a sealed copy, and any folder you open is brought in with its remotes quarantined — so nothing phones home. No cloud upload, no training on your data. Local-network teamwork and an optional relay are opt-in, per workspace.
Agents only ever work on a copy; nothing changes your real files until you approve it, and every step is logged and reversible. Because nothing leaves your machine without an explicit choice, it fits confidentiality and data-protection duties — for example §43e BRAO and DSGVO for legal work — rather than working around them.
Claude, GPT, and local models — your choice, per run. Every run records which model produced what, so the work stays auditable.
Three packs are live — Project Management, Law Firm, and Pharmacy — with more coming. They all run on the same engine, so if your work is documents, deadlines, and review, it fits.
Light onboarding: point it at a folder or an export and you’re working. We set up your first workspace with you and shape the playbooks around how your team works. No lock-in; pricing at launch is a flat per-seat plan, not usage-based billing.
The other option is the status quo — the busywork stays manual, and the only AI shortcut on offer sends your work to someone else’s cloud.
You see every change before it’s real, every claim shows its source, and your files never leave your computer.
Free while in beta · flat per-seat at launch, never per-use metering.
Private beta · packs for project, legal & pharmacy teams · runs on your computer, no cloud required.