How Task Machine works
These docs explain how humans and agents run recurring work together: how you set a job up, how agents run it on your own machines, and how every decision comes back to you with a history you can read.
Start with the model
See how tasks, agents, and workflows fit together, and where your decisions come in.
Get set up
Go from sign-in to agents ready to work: connect a machine, install a playbook, and run the first job.
Follow the papertrail
Use tasks as the visible history for comments, assignments, approvals, and runs.
Start Here
Welcome
What Task Machine is and the shortest path to a working setup.
Getting started
Install the tama CLI, connect your first machine, and get your first agent reply, end to end.
How Task Machine works
How the three-surface workflow (chat, inbox, tasks) runs work humans and agents share.
Onboarding
The guided path from a new account to a workspace with a goal, playbook, autonomy defaults, and selected plan.
Plans and trials
Choose how each workspace runs, understand the seven-day trial, and keep model spending predictable.
Private beta
What to expect from Task Machine during the private beta, and how to get the most out of it.
Build Your Company
Build a company from zero
The founder path through these docs: five stages from no idea to a running company, with agents carrying the recurring work at every step.
Start with a wedge you understand
How to choose a first business idea in the agent era: filter for problems you know, audiences you can reach, and the narrowest version someone would pay for.
Validate the idea before you build it
How to test whether people will pay for an idea before writing code, and how to define the evidence that would kill it.
Get your first ten customers
How to win the first customers through direct conversations and channels that work with zero audience.
Put your operations on rails
How to keep fixed costs near zero, write down every recurring process, and turn the written processes into supervised workflows.
Scale what already works
How to grow by multiplying proven loops — cheaper first, then wider, then adjacent — and when to hire a human instead of adding an agent.
Run an AI Company
Run a company with agents
The operating model for running a company's recurring work with AI agents while every consequential decision stays yours.
Where agents fit first
Which recurring work to hand to agents first, and which work to keep supervised while trust builds.
Control without becoming the bottleneck
How to keep every consequential decision yours without personally approving everything agents do.
Autonomy levels in practice
How to raise an agent's autonomy step by step as its work proves itself, and what each level feels like day to day.
Budgets and real money
How token and money budgets cap agent spend at any scope, with alerts before the cap and a hard pause at it.
Deciding when to trust agent output
How verifier steps, run history, and approval records turn trust in agent work into an evidence-based decision.
From one playbook to an operation
The growth arc from one installed playbook to a set of running workflows that carry the company's recurring work.
Set Up Your Workspace
Workspaces
The boundary that holds one company or client and everything its work touches.
Members and roles
Who belongs to a workspace, human or agent, and what each is allowed to do.
Teams
Groups of workspace members, people and agents together, for ownership and routing.
Projects
The containers that group related tasks and give every task a readable identifier.
Task intake
Let outside systems file tasks into a project through a per-project secret token.
Goals
Workspace outcomes that gather tasks across projects under a responsible lead.
Labels
The project-scoped tags that classify tasks so you can filter and scan them.
The Three Surfaces
Chat
The chat surface where you reason with an agent and that thinking fans out into work.
Inbox
Where everything that needs your judgment comes back to you.
Tasks
The execution surface and the papertrail where you steer work and read its history.
Comments and mentions
How discussion on a task draws the right person or agent in through the timeline.
Command center
Search, jump, create, and act across the current workspace from one shell surface.
Put Agents to Work
Agents
Agents are workspace members you assign work to and bound the same way you bound people.
Agent profiles
The profile is where you say how an agent works and where its judgment ends and yours begins.
The agent loop
How a trigger becomes an agent run on your machine, and how the run's results become product state that drives the next one.
Worker machines
Agents run on computers you control, through Task Machine on your computer reporting the coding tools it finds.
Worker tool support
How each coding tool's own command and output become work Task Machine can run and read.
Supported coding tools
The coding tools Task Machine supports, and how a machine offers them to agents.
Connectors
Connectors let your agents act in the services you already run — each one backed by a real Model Context Protocol server.
Knowledge & Context
Make Work Repeatable
Workflow builder
Author a recurring process once as a graph of agent, branch, approval, and document steps.
Workflow execution
How a workflow run executes under a task — node by node, with verifier gates and human approvals.
Playbooks
Install a ready-made bundle of agents, goals, workflows, and schedules that solve one recurring job.
Stay in Control
Constitution
The two-layer policy every Task Machine agent follows: a platform baseline you cannot weaken and workspace rules you add on top, reviewed before they take effect and enforced as agents work.
Approvals and verifier gates
The checkpoints a workflow carries — a human approval, a question for a person, and a verifier that checks an agent's output.
Budgets and retries
The two bounds on repeated agent work — budgets that cap spend across scopes, retries that cap repeated attempts.
Vault
The workspace vault holds logins, API keys, and secrets — only the password or secret is sealed, and it is write-only by design.
Reference
CLI command reference
Every tama command, its arguments, and what it does.
Permissions reference
The permission keys Task Machine enforces and the roles that carry them.
Markdown reference
The markdown and reference syntax Task Machine understands in descriptions, comments, and documents.
Troubleshooting
Fixes for the setup and worker problems you are most likely to hit.