Takuto Core overview
Takuto Core is a self-hosted AI coding pipeline that works at your pace. Point it at Jira or GitHub Issues and let it run autonomously — picking up tickets, creating a branch, writing code, running tests, and opening a pull request — or stay in the driver’s seat and trigger each phase yourself from the dashboard. Every workflow runs inside its own Docker container, behind an egress firewall, on hardware you control.
Beta. Takuto Core is in active beta. It is usable today, but expect rough edges — and please send feedback. Every report helps shape where the project goes.
What you can do with it
- Fully automated mode — connect Jira or GitHub Issues and Takuto polls automatically: it picks up “To Do” tickets, runs the full pipeline (worktree → install → implement → lint/tests → PR), and moves on to the next one.
- Manual mode, your pace — add any ticket or task to the dashboard yourself, refine its description with AI assistance before the agent ever sees it, then trigger each workflow phase when you’re ready.
- Mix both — auto-pick routine work while curating the tricky tickets by hand.
- Run tickets in parallel — each workflow gets its own git worktree and isolated environment; you choose the concurrency.
- Monitor in real time — a live dashboard streams terminal output per workflow and lets you pause, resume, retry, or inspect any run.
- Define your own pipeline — chain phases (implement → review → test → PR → merge base) with dependencies, edited in the dashboard’s Workflows tab. The bundled TOML files seed the defaults for a new workspace.
- Work without a ticketing system — paste a description into the dashboard and Takuto treats it as a workflow. No Jira account required.
How it differs from an IDE assistant
| IDE assistant (Copilot, inline Cursor) | Takuto Core | |
|---|---|---|
| Where it runs | In your editor, on your machine | In Docker, on any machine or server |
| Supervision | You approve each step | Autonomous or manual-trigger — your choice |
| Ticketing | None | Jira, GitHub Issues, or standalone |
| Pipeline | Single prompt | Multi-step TOML: implement, review, test, PR |
| Concurrency | One task at a time | Multiple tickets in parallel |
| Security boundary | Full internet access | Egress firewall — only approved hosts reachable |
| Team deployment | Per-developer only | Self-host on a server; shared dashboard |
| Persistence | Ends when you close the editor | Survives restarts; paused workflows resume |
Two ways to run it
There are two supported paths. Pick the one that matches how much control you need.
1. The Takuto CLI (recommended for getting started)
The companion Takuto CLI generates your config and orchestrates Docker
Compose for you. From an empty directory you get to a running dashboard in minutes: install
the CLI, run takuto setup (it creates a .takuto/ folder in the current directory), then
takuto start to bring it up and open the dashboard — where you finish setup (admin
account, AI provider, GitHub). This is the fastest way to try Takuto Core on a single
project.
→ Start with the Quick start.
2. Build your own container from Takuto Core
If you want full control — a server deployment, a custom image, a reverse proxy in
front — clone the Takuto Core engine from https://github.com/takuto-team/takuto-core and run it
with Docker Compose directly. You manage config.toml, the build, and the deployment
yourself.
→ See Install Takuto Core.
Both paths run the same Takuto Core engine and the same dashboard. The CLI is just a convenience layer over the Compose setup.
Prerequisites
Before you start, gather these:
- Docker (or Podman) with
docker compose— Docker 24+ / Podman 4+ recommended. The Takuto CLI auto-detects whichever you have. - RAM: ≥ 8 GiB for a single workflow; ≥ 12 GiB on macOS with Podman (the Podman
VM needs its own share). Tune
[general] max_concurrent_workflowsto your machine. - Disk: ≥ 30 GiB free. Worktrees, npm/cargo caches, mise toolchains, and the optional Docker-in-Docker storage layer all live in Docker volumes.
- A GitHub account and token — a fine-grained personal access token (PAT) or a
configured GitHub App. See Configuration →
[github]. - An AI provider account — Claude Code (Anthropic API key, Pro/Max OAuth, or a corporate proxy), Cursor Agent, Codex, or a self-hosted model via OpenCode.
- (Optional) Atlassian access — only when
ticketing_system = "jira".
A Linux host is recommended for server deployments; macOS works well for local use.
Licensing at a glance
- Takuto Core (the engine) is source-available under the FSL-1.1-ALv2 (it converts to Apache-2.0 two years after each release). Free for non-competing use; if the terms don’t fit, reach out at morphet.contact@gmail.com.
- The Takuto CLI is MIT licensed.
See the FAQ for details on licensing, cost, privacy, and beta status.