tasskz
Download

Your AI agents do the task You approve

tasskz is a platform that runs AI agents for you. Create as many tasks as you like and run each one with the agent of your choice (Claude Code, Copilot, Codex, Gemini and more). Come back only to review, ask for changes and approve. Everything runs on your own computer.

    Works with the agents you already use

    Use the agents and models you already know, on the plan you already pay for. No paying twice for the same model, and you can put more than one agent to work at the same time.

    • Claude CodeAnthropic
    • GitHub CopilotGitHub
    • CodexOpenAI
    • GeminiGoogle

    From idea to approved result in four steps

    You decide what gets done and what gets approved. The agent does everything in between.

    1. Backlog
      01

      Create the task and pick the agent

      Write the main prompt and attach context: notes, links, local folders and files. Pick the agent, model and effort, or leave it on Auto and tasskz uses the first available agent on your priority list.

    2. In Progress
      02

      Watch it run

      The task runs in the background. The card shows a timer and what the agent is doing right now, and moves to Review by itself when the run ends.

    3. Review
      03

      Review it on your time

      If you turned on human review when creating the task, you are notified when the work is ready for you. The task waits in Review until you act: open the outputs, ask for changes in the conversation or approve.

    4. Done
      04

      Approved and versioned

      When you approve, the task moves to Done and the approved version is recorded in git with a commit and a tag. Outputs and history are kept, and you can reopen the task at any time.

    Everything you need in one platform

    Built for people who run more than one agent task at a time.

    Context that travels with the task

    Add all the context you have for the agent to use: notes, links, local folders and files, gathered in one place for the run.

    Dictate your prompts

    You don’t even have to type if you don’t want to. Voice transcription and AI prompt improvement do it for you.

    Live execution

    A timer and the current activity on each card. Open the task for the full timeline of every tool call.

    Parallel runs with a queue

    Two sessions at once by default, configurable up to unlimited. Extra tasks wait in the queue and start on their own.

    A conversation per task

    Reply to ask for changes and the agent continues in the same session. When the agent has a question, the card asks for your attention.

    Outputs versioned with Git

    Each run is committed to the task workspace. Open any version, see which files a run added or changed, download a ZIP.

    Review the work whenever you want

    Nothing is finished until you say so. Tasks stop in Review and wait for your approval. You can turn this off per task.

    Know when it needs you

    A notification bell and native Windows notifications when a task is waiting for review, has a question or failed.

    Every detail of each run, and the full history

    Each run keeps the tool timeline, raw log, CLI errors, sent prompt and metadata: model, effort, cost, tokens and duration. The History tab records everything that happened to the task.

    9 languages, dark and light

    English, Português, Español, Deutsch, Italiano, Français, 简体中文, हिन्दी and Русский.

    The right agent for each task. Or let Auto mode choose.

    You are free to choose the agent, model and effort of each task.

    One agent per task
    Claude Code for one task, Copilot, Codex or Gemini for another. Each task remembers its own choice.
    Auto (by priority)
    Drag your agents into a priority order. When a run starts, tasskz uses the first one that is installed, signed in and available, and shows you which one it chose and why.
    Installed and signed in from the app
    Install the agent CLIs and sign in from the settings screen. A CLI that is already on your machine is detected and used.
    Your accounts, your limits
    tasskz does not include or resell model access. It runs the CLIs with your own Claude, GitHub, OpenAI or Google accounts.

    Runs on your machine. You stay in control.

    Here is exactly what stays local and what does not.

    Stays on your PC

    localhost
    • The app and its database, your tasks, context, outputs and logs.
    • The workspace, files and git history of every task.
    • Agent API keys and credentials.
    • Voice dictation: the audio is transcribed locally and discarded.

    Goes to the network

    HTTPS
    • What the agent reads and writes is sent to the provider of the agent you chose (Anthropic, GitHub, OpenAI, Google), by its own CLI
    • One-time downloads: agent CLIs and voice models, over HTTPS from fixed domains
    Local only by default
    The app runs only on your local machine. No external access.
    Password protected
    You set a password on first run and choose how long your session lasts before it expires.
    You control what the agent can do
    Pick the execution profile: Auto, Documental or Total. With Documental, the agent reads and writes files but does not run commands.

    Up and running in minutes

    1. Download and install

      Per-user installer, no administrator rights. Git and the database are included: there is nothing else to install first.

    2. Connect one or more agents

      The setup wizard installs the agents you choose (Claude Code, GitHub Copilot, Codex or Gemini) and signs you in with your own account.

    3. Create your first task

      Write the prompt, press "Create task and start" and watch the card move.

    Download for Windows

    Requirements: Windows (x64) and an account with access to Claude Code, GitHub Copilot, Codex or Gemini (any one of them). Node.js is needed only if you use GitHub Copilot.

    Frequently asked questions

    What is tasskz?

    A desktop app that automates your tasks with AI agents on a Kanban board. Each task has its own prompt, context and workspace. An agent CLI (Claude Code or GitHub Copilot) runs the task in the background, and you review and approve the result.

    How is this different from running Claude Code in a terminal?

    A terminal gives you one conversation at a time. tasskz gives you a board: several tasks running or queued, a clear list of what is waiting for you, a full history of each run, and outputs versioned with git. If you need the terminal, "Continue in terminal" opens the same session there. On top of that, you get one unified interface to work on, review and organize all your tasks.

    Who is it for?

    It is made for everyone who wants to organize and automate their tasks! tasskz can be used by technical people and to write code, but it is not limited to that: it solves everyday problems by producing content, analyses, reports, presentations and much more.

    Which agents are supported?

    Claude Code and GitHub Copilot today. Support for Codex, Gemini CLI and others is coming soon.

    Can I use more than one agent at the same time?

    Yes. Each task has its own agent, and tasks run in parallel: two sessions at once by default, and you can raise the limit or make it unlimited. Extra tasks wait in a queue. Within one task there is one agent at a time.

    Do I need an API key or a subscription?

    You need your own access to at least one agent: a Claude account for Claude Code, or a GitHub account with Copilot. You sign in from the app. tasskz does not include or bill model usage.

    Does my data leave my machine?

    tasskz itself has no cloud service: the app, database, files and history stay on your PC. The agents do talk to their providers: what an agent reads and writes during a task is sent to Anthropic or GitHub by its CLI, under your account and their terms.

    What can the agent do without my approval?

    That depends on the execution profile of the task. Documental lets the agent read, search and write files, with no commands. Auto lets the agent decide within the task workspace and denies what it considers risky. Total runs everything without prompts. In all cases the task stops in Review before it is marked as done, unless you turn Human Review off.

    How does the review work?

    When a run ends, the card moves to Review. You read the agent summary and open the review panel to evaluate the generated files and ask for changes, which runs the agent again in the same session. Approve moves the task to Done and tags the approved version in git.

    What are the requirements?

    Windows (x64). The database and git come with the installer. Node.js is needed only for agents installed through npm, such as GitHub Copilot. Voice dictation is optional and downloads its models on first use. You need an account with one of the AI agents tasskz supports, and tasskz itself lets you install those agents from the app.

    How do I download and install it?

    Download the installer, run it and choose a folder. On first run you set a password, and a setup wizard helps you install and sign in to one or more agents. Windows may show a SmartScreen or security warning, but it is safe to continue.

    Does it run on macOS or Linux?

    Not today. tasskz is a Windows (x64) app. Support for these operating systems is coming soon.

    Get all your tasks done in one place.

    Download tasskz and run your first task today.