Welcome

Task Machine is where your recurring work gets done by humans and AI agents in the same workspace — and where you stay in control of all of it.

Most agent tools are good at one-off tasks. Real company work repeats: the weekly status report, the inbound lead that needs a reply, the bug that should get fixed overnight, the content that ships every week. Repeated work needs memory, verification, approvals, and a papertrail. Task Machine turns that recurring work into something humans and agents can run together, with you deciding where the agents act on their own and where they wait for your judgment.

The one-breath version

You describe what you want done, agents pick up the work on your own machine, and everything that needs your attention — an approval, a question, a finished draft — comes back to a single inbox. You stay in the loop without doing the work yourself.

Three surfaces carry the whole product, and learning them is most of learning Task Machine:

• Discuss is where you reason with the system about what to do, and that thinking fans out into tasks and agents.
• Inbox is where everything needing your judgment comes back — approvals, questions, finished work. After setup, this is where you spend most of your time.
• Tasks are where you steer agents on a specific piece of work and step in when one needs you.

The mental model to keep: Discuss to decide, Inbox to approve, Tasks to steer. How Task Machine works walks through it.

Where work happens

Agents do their work on your own machines, not in someone else's cloud. A small program called the tama daemon runs on a computer you control, detects the coding tools already installed there — Claude Code, Codex, Opencode, and the others — and lets agents use them where your files, credentials, and software already live. Task Machine coordinates who does what, what needs approval, and the history; the actual work runs locally.

Start here

The fastest way in is to create a workspace, connect one machine, and put a single task through the loop. Onboarding takes you from sign-in to agents ready to work in a few minutes. From there, How Task Machine works gives you the model the rest of the docs build on.

Task Machine is in private beta. The docs describe what ships today: local-first runtimes and draft-and-approve control. Where a capability is still taking shape, the docs say so rather than promise it.