Tack
EnglishEspañol
Download

Open source · MIT · v0.1.0-beta.9

A project board that hands its items to your coding agent.

Press ▶ on a card, pick Claude Code, Codex, docket or opencode, and it runs. What happened stays on the item. Self-hosted: one binary, one SQLite file, no account.

Download the desktop appHow a run goes

Linux (.AppImage, .deb) · macOS (.dmg, Apple Silicon and Intel) · Windows (.msi)

  • Desktop app
  • Four agent harnesses
  • One SQLite file
  • No account
The Tack board: Kanban columns with WIP limits, a run button on every card, and a banner offering to let this board run its items with an agent.

What it is

Your plan and your agent, in one place.

The run stays on the card.

Events, questions, files and cost stay on the item after the agent is done, not in a chat log you have to scroll.

A run survives a crash.

Kill a run halfway through and Tack neither loses it nor starts it again behind your back. It stops and asks you.

Your keys stay with you.

The board never holds a model credential. The runner that does lives where your code already is.

Measured, never guessed.

Cost and tokens show what the agent reported, or say “Not measured”. Never an estimate, never a silent 0.

18.5 MiB
one binary
~18 MiB
of memory at rest
4
agent harnesses
0
accounts, database servers, Docker or cloud

Agent runs

Plan it on the board. The agent does the rest.

Three steps, the same whichever agent you pick.

  1. Plan it on the boardCreate the item and pick who runs it.
  2. A runner picks it upIt checks out a clean workspace and starts the agent with its own login or a key you gave it.
  3. Watch it on the itemProgress arrives live. The finished run stays in the item’s history.
Four harnesses, one flowClaude CodeCodexdocketopencode

You choose how it is paid for.

Use the agent’s own subscription and Tack never sees the credential, or give the runner an API key for an endpoint. The model list comes live from that endpoint.

It can ask first.

With approvals set to “Ask me”, Claude Code, Codex and opencode pause before acting and wait in the item’s decision inbox.

The Agents page gets you started.

From installed binary to a finished test run on one page: turn execution on, see which agents this machine has, check their login, pick a default model, try it.

Without agents too

A full project manager, with or without agents.

Everything here works the moment Tack starts, with no agent set up.

  • Six views: board, list, table, calendar, timeline and overview.
  • Your workflow: Scrum, Kanban or phases, with WIP limits.
  • Rename Task, Sprint or Backlog per project. The UI, CLI and API all use your terms.
  • Dependencies that matter: cycles are rejected, and an item waiting on unfinished work is not handed to an agent.
  • Export to JSON, YAML or CSV. Import from GitHub Issues or Linear. Back up locally or to S3-compatible storage.
Timeline: a dependency-aware Gantt view. The same dependency graph decides which items an agent may pick up.
The timeline. The same dependency graph decides which items an agent may pick up.

Install

Download the app and open it.

Desktop app

Open it and the board opens in its own window. Closing the window does not stop the work.

  • Linux.AppImage or .deb
  • macOS.dmg, Apple Silicon and Intel
  • Windows.msi
Download the latest release

Binaries and the desktop app are not code-signed yet. On macOS, right-click → Open the first time. On Windows, choose More info → Run anyway.

And tack serve

The same binary and the same data, run from a terminal. It is for a machine with no desktop, a remote or shared host, or automation through the CLI and REST API. For a shared server, attach a separate tack-runner instead of using --with-runner, which only runs on localhost.

binary
curl -fsSL https://raw.githubusercontent.com/yielab/tack/main/install.sh | sh
tack serve --with-runner

Then open http://localhost:3210. The installer checks every download against the release’s SHA256SUMS. Your data is tack.db plus the storage/ folder. Back up both.

Read the quick start →

Straight answers

What Tack is not.

Roadmap →
  • Not multi-user. There are no accounts: one optional shared token, no per-user permissions.
  • Not live sync. There is one SQLite writer, and S3 backup copies snapshots.
  • Not available offline, and there is no native mobile app.
  • Not always exact on cost. Imported cost values may be estimates; Tack’s own measurements always say where they came from.
  • Not code-signed yet. Expect a first-run warning on macOS and Windows.
  • Not finished. It is a public beta.

Put your plan and your agent on the same board.

Open source under MIT. Download it, run it on your own machine, and keep every run on the item it belongs to.

Download the desktop appGitHub