Changelog

Technical builders ship the product, then watch marketing, outreach, support, content, and ops pile up — the work a first employee would do, except hiring is off the table. Task Machine is the inbox-first operating layer for that recurring work, run by humans and agents together, and it is now open in private beta.

You direct an agent team, review what needs your judgment, and let the rest run — with control where you want it.

• Three surfaces, one workflow — Chat to set strategy and fan work out, the inbox to approve and review, and tasks for the back-and-forth on a specific piece of work. Chat to direct, inbox to approve, tasks to dig in.
• Templates and guided onboarding — start from a job-specific bundle for your work instead of configuring agents, workflows, and goals by hand. Onboarding installs the bundle, connects a runtime, and lands a real deliverable in your inbox.
• Deterministic workflows — recurring work runs as explicit, repeatable graphs with branch conditions, verifiers, approval steps, retries, and step-level logs, so a run does the same thing every time and you can read what happened.
• An agent loop you steer — assign a task or comment on one to put an agent to work; team leads route shared work; guardrails keep loops bounded; and every run ends with a summary in the timeline.
• Proposals and approvals — agents propose tasks, workflows, skills, and team changes into the inbox, where you approve or reject with a reason. Nothing proposed acts until you say so.
• Budgets and usage analytics — set per-agent budgets, handle increase requests from the inbox, and see where credits and time go across agents and workflows.
• Connectors — wire agents to the tools and accounts you already use, and see what each template works with before you install it.
• Knowledge and memory — internal references, a shared document picker, and durable agent memory carry context between runs instead of losing it to terminal history.
• Notifications — browser push for inbox-worthy events and a daily digest email that pulls together what your agents did and what now needs you.

v1 runs on local runtimes: install the tama daemon and bring your own provider access, and agents execute on your own machine. The control story is draft-and-approve — agents do the work, you review it from one inbox like a manager. Cloud runtimes come later.

To get started, pick a template that matches your work, connect a runtime, and approve your first agent's deliverable from the inbox.

Knowing whether your agents are actually clearing work used to mean scanning the task list and counting. Tasks, goals, and projects now carry a progress breakdown and a work-item burndown, so how much is done, what is still open, and whether the trend is heading the right way reads at a glance.

What changed

• The Tasks page splits into two tabs. Tasks keeps the board and list focused on the work, and a new Analytics tab holds the progress and burndown cards so the numbers stay out of the way until you want them.
• Goals and projects open on their tasks as the overview, each with a Progress card — completed, started, unstarted, and backlog, with cancelled items excluded — and a burndown that climbs as work is created and falls as it closes, toward the goal's due date when one is set.
• A project's linked goals moved into its settings, so opening a project lands you straight on its work.

Reading the cards

Progress answers how close a goal or project is to done; the burndown answers whether the open work is actually going down. A line that rises faster than it falls is the early signal to step in before the backlog gets away from you.