Two paths. Self-host: clone the repo, bring your own keys, run it on your infrastructure - you're in a working workspace in a few minutes. Or use the managed cloud and skip setup entirely. Either way, the first move is the same: create a workspace, point an MCP-speaking agent at it, and drop it a comment to act on.

Teams building products and the founders running them. Today's early users are AI native PMs who want to put agents on real work rather than in sidebars, plus GTM leaders treating their pipeline as something agents can actively operate alongside.

Free to self-host and to use the open-source build with your own keys. We offer a paid version with hosting and a marketplace of closed loops.

Self-hosted: your infrastructure, your rules, no data leaves your network. Managed cloud: EU-resident, no training on your data, retention only as you configure. Auth via standard OAuth providers. Every agent action is logged as an event on the object it touched fully auditable.

You choose. Agents are scoped to the objects, tools, and workspaces you grant them. Every read and write is a logged event you can inspect. Skills (the layer that defines agent behaviour) can require an explicit human @mention before high-stakes actions that's how our own setup works. Pause or stop any agent session from the UI at any time.

Two founders are running their own company on it (us). Design partners are using it in production. It is not v1.0 in the "five years of polish" sense. It is in the "we ship every day and you can read the changes on GitHub" sense.

Model Context Protocol is the open spec for how agents describe tools, list resources, and stream work. We built Maskin against MCP from day one. No proprietary agent SDK, no vendor lock-in. If your agent speaks MCP, it works.
In practice that's Claude, Claude Code, Cursor, Codex, and any custom agent you build against the spec.

Those tools wire agent into one app. Maskin is the shared surface those agents are missing. Drop any MCP-speaking agent in and it reads the workspace as a peer same objects, same comments, no glue code. Open source under your control rather than a closed SaaS sidebar.

Assemble a team of specialized agents in a workspace where humans and agents work against the same objects: tasks, insights, bets, knowledge and threads.
Raw signal comes in from your sources, agents cluster it into insights and propose them as bets, you decide what's worth it. From there, agents do the work and you review the result.