TasmaThe local workflow manager for coding agents GitHub

Every step gets a fresh agent. The task remembers the work.

Free and open source. macOS 14.5 or later.

Runs your own Claude Code, on the plan you already have. No API key.

A task, and the chat of the step that runs now.

A fresh chat that knows the whole story.

In Tasma, every step starts in a fresh chat, with one job and a clear context.

Before it starts, it reads the task: what you asked for, and what each step before it did.

When it finishes, it writes its report on the task, for the next one.

The task stays, and grows

The task Canary rollouts with automatic rollback, in progress, on the step dev:implement.
The chats of this task, one for each step it ran: research, implement, review, and implement again.
The goal of the task, written before any agent started.

Comments the reports of the steps

  • Dev notes dev:implement
  • Review #1 dev:review
  • Dev notes #2 dev:implement

The chat new at every step

Chat 4 of the task, for the step dev:implement.
2 tool callsReview #1: two findings. The rollback leaves a timer armed, and the timeline does not show the reason of a rollback.I wrote both on the task.
An empty chat. It starts with the task.
The first lines of the new chat: the agent already works on the findings that the review step wrote on the task.
The end of the chat: both findings are fixed, all tests pass, and the agent added its notes to the task.
The box under the chat, where you write to the agent.

You define the workflow.

A workflow is a list of steps. Each step has its own instructions and its owner: you with an agent, or agents alone.

When a step is done, the agent can start the next one in a fresh chat. A step that is yours waits for you.

Tasma comes with one workflow, for development. Change it, or write your own.

Comes with Tasma.

    TASMA-891

    Add SSO login

    Comments on the task

      A step that is yours. It waits for you. A step for agents alone. A drawing of the idea, not a screen of the app.

      Tasks and chats in one app.

      A board for your tasks.

      Tasks are Markdown files on your Mac. Each card shows its step.

      The Tasma board with the columns To Do, In Progress and Review. Each card that is in work shows its step, such as dev:implement or user:review, and how far it is along its workflow.

      All your chats on one screen.

      Run several tasks at once. See which agents are working and which are waiting for you.

      One list with the chats of every task and project. Each row names the task, its step and whether the agent needs you.
      The open chat: the dev:implement step of Canary rollouts with automatic rollback.
      The agent's last messages in the open chat, and under them the box where you write to the agent.

      The list. Every chat, of every task.The open chat. You read what the agent did, and you write back.

      The Tasma Method

      Six principles for work with agents. They fit any workflow.

      1. Write the task first.

        The task is the starting point. Every step begins there and adds its result to it.

      2. Split the work into steps.

        One step, one job: research, build, test. A small step is easy to check and easy to fix.

      3. Start every step fresh.

        A fresh agent questions what a long chat takes as settled. It finds what the first one missed.

      4. End every step with a report.

        The agent writes what it did and why. The next agent reads it, and so can you.

      5. Fix the workflow, not only the result.

        When a result is wrong, change the step that let it through. Every later task gets the fix.

      6. Decide which steps need you.

        Some steps need your judgment. Agents run the rest.

      Tasma is built with Tasma.

      3,000 lines a day. One developer.

      Give your next task a fresh start.

      Runs your own Claude Code, on the plan you already have. No API key.

      Free and open source, under the MIT license. macOS 14.5 or later.

      The app guides you through the setup.