Private beta

Task Machine is in private beta, so it helps to know what to expect before you lean on it for real work. The product is usable today, but it is early: some flows are unpolished, and agents need more supervision than they will once things settle. This page sets that context so the rough spots are expected, not surprising.

Agents run on your machine, with cloud runtimes coming

Today, agents do their work on machines you connect — a computer you control runs the work using the tools installed there, and Task Machine coordinates the rest. Hosted cloud runtimes are on the way, so this setup gets simpler over time; until they land, a workspace needs at least one connected machine before any agent can do anything. Onboarding walks you through connecting one, and machines and runtimes covers it in depth.

Set up tools and credentials where agents run

Because the work happens on your machine, an agent can only use what is available there. The coding tools it runs on, web and browser access for research, and any API keys or credentials a task needs all have to be configured locally, on the machine the agent uses — not in the Task Machine app. If a task depends on something an agent cannot reach from that computer, it will stall or work around the gap, so set up the local environment for the jobs you intend to run.

Expect rough edges

This is a beta, and it shows in places: some screens are unfinished, some interactions are awkward, and behavior will change between updates. None of it should lose your work — the task history records what happened — but treat the experience as still taking shape rather than final.

Watch the agent loop

Agent runs are not fully reliable yet. An agent can stall, drift from what you intended, or finish with work that needs correcting, so keep an eye on active tasks rather than leaving them unattended. The task timeline and its run transcript show exactly what an agent did, which is where you catch a run going wrong and step in — redirect it with a comment, or retry. Set autonomy no higher than you are willing to supervise during the beta: the more an agent does without asking, the more closely it is worth watching while the loop matures.

Tell us what breaks

The beta is where the rough edges get found, so feedback is the point. When something is confusing, missing, or broken, email support@taskmachine.io or message @fabian_builds on X — concrete reports about a specific surface or task move the product fastest. When an agent does the wrong thing, the task timeline and its run transcript are the best evidence to share.

From here

If you have not set up a workspace yet, Onboarding is the guided path from sign-in to agents ready to work. To understand the model the rest of the docs build on, read How Task Machine works.