Skip to main content

ntasker

Put your coding agent on a kanban board.

Drop your tasks on the board, hit run, and Claude Code, OpenCode or Pi picks one up -- in a real terminal embedded in the page, in that task's project directory, already briefed on the task. A queue works through the rest unattended, one task per project, while you do something else. Runs on your own machine: one Python package, a SQLite file, no account, no build step, nothing leaves the box.

ntasker kanban board with the projects sidebar and the queue panel

Quickstart

uv tool install ntasker                  # install from PyPI
ntasker service install --auto-update    # run as a service + daily auto-update
ntasker config set projects_dir ~/Projekte

Open http://127.0.0.1:8766. The service creates the database on first start, restarts on crash and keeps itself up to date. On Linux, run loginctl enable-linger $USER once so it survives logout. No supervisor wanted? ntasker serve runs it in the foreground until you close it.

Then teach your agent about it:

ntasker agent install claude     # or: opencode, pi

Why

Three things a plain to-do list cannot do.

The board is your agent's memory

The installed skill and /task command let the agent read and drive the tracker -- no copy-paste, no re-explaining:

  • "What should I work on next?" -- it grabs the open tasks for your current project folder and ranks them by urgency.
  • /task 34 -- pulls #34 into the session (title, description, tags), flips it to in progress, and warns you if you are sitting in the wrong project.
  • "Add a todo: ..." -- it files the task for you; drop a #34 anywhere later and it knows which task you mean.
  • Finished an assigned task? It moves the task to Review for you to sign off. It never closes, deletes or archives anything on its own.

Run a task without leaving the board

Every task row has a run button showing that task's agent logo. It opens the genuine TUI, embedded in the page via xterm.js, running in the task's project directory and seeded with the task. You answer the agent's questions, approve its tool prompts and interrupt it exactly as in a terminal -- same CLI, same CLAUDE.md, same skills, MCP and permissions.

Sessions run in the background (the button shows a spinner; re-opening reattaches to the live session), and marking a task done ends its session. An open task whose session already ended gets a resume button next to Run, which continues that conversation instead of starting over. The button only appears when the agent's CLI resolves and a POSIX pseudo-terminal is available. See docs/claude-runs.md.

Interactive Claude Code session embedded in the ntasker web UI

The queue keeps going when you stop watching

Every run button puts its task at the head of its project's lane, and ntasker works through the queue one task per project at a time, taking the next one as soon as the previous is closed -- by you after review, or by the agent when the task told it to. done is the only thing that ends a session. Pause stops new starts; running tasks keep going.

The panel shows one column per project, because that is what runs in parallel. To make one task wait for another -- across projects too -- drop it on the middle of the other; the edges keep reordering. A fasttrack task commits and closes itself and hands its result to the tasks depending on it; the Run-Log collects those outcomes. Plan queue lets an agent session order the queue for you and start it. See docs/task-queue.md.

The inbox sorts your ideas

Type a raw thought into the topbar field (Ctrl+K) or ntasker in "..." and carry on. A stateless claude -p call turns it into a proposed task -- title, prompt, priority, tags and the project, chosen from a catalog of one-paragraph project summaries -- and the Inbox column shows the proposal with its candidate projects pre-ticked. One click accepts it; tick another project, or several (the rest become directory locks), and every correction becomes an example for the next triage. Nothing becomes a task until you say so.

See docs/inbox.md.

Pick your agent per task

ntasker is agent-agnostic: Claude Code, OpenCode and Pi are supported out of the box, and adding another is one plugin. Each task carries an agent and an optional model; either can fall back to a global default.

ntasker add --title "..." --agent opencode --model opus
ntasker config set default_agent opencode
ntasker agent list                             # CLI availability + integration status per agent

Full reference incl. per-agent binary paths: docs/agents.md.

Plugins

Optional features ship as plugins you can switch off individually (/settings -> Plugins):

Plugin What it adds
task_context Attach files, notes, personas, skills and MCP servers to a task, handed to the agent in its briefing (docs)
workspace Team (Claude Code subagents), skills, knowledge base and documents on a /workspace page (docs)
voice Dictate task descriptions with local speech recognition; opt-in via ntasker enable voice (docs)

A disabled plugin's routes 404, its agent is neither listed nor resolvable, and its data stays intact. Contract and slots: docs/plugins.md.

Documentation

Topic
Configuration Settings, DB path, projects_dir, language, vendor assets, why you must not expose the port
CLI reference Every subcommand and flag
HTTP API Endpoints, SQLite schema, design notes
Agents The agent registry, per-task model, skill installation
Agent runs How an embedded session is spawned and reattached
Task queue Queue semantics, dependencies, fasttrack, run log
Directory locks Keeping two agents out of the same working directory
Kanban view Board vs. list view, drag-and-drop, keyboard shortcuts
Inbox Raw notes triaged into task proposals by a stateless claude -p call

| Projects | Sidebar tree, project families, misc project | | Service | systemd / launchd, auto-update, uninstall | | Development | Repo setup, smoke test, translations |

Coming from the drfoehn fork? See docs/migrating-from-fork.md.

Stack

FastAPI + uvicorn on the Python stdlib sqlite3 -- no ORM, no migration files. The frontend is HTML + AlpineJS + Tabler.io loaded from jsDelivr with pinned SRI hashes, so the wheel stays under 100 KB and there is no build step; an offline mode is one command away. Requires Python 3.12+.

Binds to 127.0.0.1:8766 and has no authentication -- it is a personal local tool, not a multi-user service.

License

AGPL-3.0-or-later. The Affero clause means: if you run a modified version as a network service, you must offer the modified source to its users. For local single-user use this has no practical impact.

Changelog: CHANGELOG.md -- issues and source: https://github.com/nerdocs/ntasker

If it saves you an afternoon, you can buy me a coffee.

Release files for ntasker 3.11.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for ntasker 3.11.0
File Size Uploaded
ntasker-3.11.0.tar.gz 967.9 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for ntasker 3.11.0
File Interpreter ABI Platform
ntasker-3.11.0-py3-none-any.whl Python 3 none any Details

Total release size: 1.5 MB

Release files / ntasker-3.11.0.tar.gz

Download URL ntasker-3.11.0.tar.gz
Size 967.9 kB
Tags Source
SHA-256 checksum
How to use checksums
c9ead848a376c3a91167142433c191c0f994114a410fc3b38f06477b575817c7
BLAKE2b-256 checksum
How to use checksums
b6e54df733dc047e0b5458f7cc61e39c9f0826e6d0917e8265c0539d2fb113a3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 25, 2026.

Transparency log

Release files / ntasker-3.11.0-py3-none-any.whl

Download URL ntasker-3.11.0-py3-none-any.whl
Size 562.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
b223d4e3a3a3b68203b6371dbe9a36e17bdcddbad0dcfbfb68569038693e0bc6
BLAKE2b-256 checksum
How to use checksums
0b787b45273e9912909db530ce3c0d7c4ce5b685576c5890052ae30995f3f07a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 25, 2026.

Transparency log

Release history Release notifications | RSS feed

This release

3.11.0 This release

2 release files

3.10.0

2 release files

3.9.0

2 release files

3.8.0

2 release files

3.7.1

2 release files

3.7.0

2 release files

3.6.2

2 release files

3.6.1

2 release files

3.6.0

2 release files

3.5.0

2 release files

3.4.1

2 release files

3.4.0

2 release files

3.3.1

2 release files

3.3.0

2 release files

3.2.2

2 release files

3.2.1

2 release files

3.2.0

2 release files

3.1.2

2 release files

3.1.1

2 release files

3.1.0

2 release files

3.0.0

2 release files

2.20.0

2 release files

2.19.5

2 release files

2.19.4

2 release files

2.19.2

2 release files

2.19.1

2 release files

2.19.0

2 release files

2.18.0

2 release files

2.17.1

2 release files

2.17.0

2 release files

2.13.0

2 release files

2.12.0

2 release files

2.11.0

2 release files

2.10.0

2 release files

2.9.0

2 release files

2.8.0

2 release files

2.7.0

2 release files

2.6.0

2 release files

2.5.3

2 release files

2.5.1

2 release files

2.3.0

2 release files

2.0.0

2 release files

1.4.0

2 release files

1.3.0

2 release files

1.2.3

2 release files

1.2.2

1 release file

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page