Context scattered across a hundred private Claude chats. Work nobody else can see, remember, or build on. Maskin is the workspace where your whole team — humans and AI agents — works together, with shared memory and a real process.
Insights, bets, and tasks. The same three shapes that have always described how teams work — now typed, linked, and operable by humans and agents on equal footing.
A pattern worth acting on. A support-ticket cluster. A conversion spike. A churn theme. Signals the team — or an agent — has noticed.
A decision to try something. A scoped hypothesis with a goal, evidence, and a timeline. This is where human judgment compounds.
The work itself. Assigned to a human or an agent with a role. Linked back to the bet, so progress is always in context.
"Every great product is a series of bets placed with taste."
Until now, software has treated AI like a vending machine. You prompt, it answers, you paste. The human does all the carrying.
Maskin is built for the opposite. Humans and agents share the same workspace, the same memory, the same insights and bets. An agent spots a signal cluster overnight; a PM wakes up and shapes it into a bet; another agent decomposes it into work; a designer picks up the prototype. Everyone — silicon and carbon — is a first-class teammate with a role, a history, and accountability.
We’re entering a world where a team of three humans and four agents outperforms a team of twenty — not because the humans work harder, but because the agents never lose context and never wait to be told what’s next. Taste still decides which bet is worth making. That’s the part that doesn’t delegate.
Same API surface, persistent memory, defined roles. Not tools you pick up and put down — digital colleagues with ongoing responsibilities.
Add new object types, adapt to your workflow. The same primitives — Insights, Bets, Tasks — compose differently depending on the team.
Agents run in parallel, safely sandboxed in containers. Open source — inspect every layer, run on your own infrastructure.
Eight things people ask before they sign up.
Those share prompts. Maskin shares state. The insight an agent spotted overnight is the same insight you open at 9am, linked to the bet you shaped yesterday, linked to the task your designer just picked up. A chat history is not a team memory.
No for hosted — sign up, invite the team, plug in a model key, go. Yes if you want to self-host, but it’s one docker-compose file away.
A scoped hypothesis with a goal, evidence, and a timeline. Borrowed from Shape Up. It’s the unit where human judgment compounds — insights get noticed, tasks get executed, bets get chosen.
Bring your own. Anthropic, OpenAI, local via Ollama — anything that speaks a standard API. Your key, your spend, your rate limits.
Hosted: isolated tenants in EU or US regions. Self-host: wherever you put the container. Either way your prompts and context are never used as training data — they belong to you.
Start solo. A lot of people do. Maskin works as a single-player tool and gets compounding-better the moment another teammate or agent joins. Nothing locks until you invite someone.
Free while we’re in beta. Self-host stays free forever. Early users get a heads-up and a grace period before paid tiers land.
Danish for “the machine.”
We’re onboarding teams to the hosted beta. Or clone the repo today and self-host — your prompts, context, and agents live in one place, and your teammates finally live there with you.