What is Routa?
Routa turns a coding request into a card on a board and routes it through specialist agents: a Backlog Refiner writes a YAML story with acceptance criteria, a Todo Orchestrator turns it into an execution plan, a Dev Crafter implements and commits, a Review Guard checks every criterion independently, and a Done Reporter writes the summary. A Blocked Resolver records blockers and sends cards back to the right lane.
Each lane distrusts the one before it and writes its evidence onto the card, so the card ends up holding the story, plan, test results, review findings and completion record. Routa does not ship its own model; you enable a provider, and it drives coding agents through ACP and other adapters, with MCP for tools and A2A for agent-to-agent calls.
It comes as a Tauri desktop app, a CLI (npm install -g routa-cli) and a self-hostable Next.js web app, backed by a Rust Axum server. The project is MIT-licensed and lives under the GitHub account of its lead developer, Phodal.
Is Routa worth it?
Routa's strongest idea is that review should be a gate with rules rather than one more opinion. Its Review Guard rejects work with missing evidence, scope creep, a dirty git tree or failing lint, and only an APPROVED verdict moves a card to Done. The Dev lane will not hand over until a commit exists and the worktree is clean. Those are the habits people try to enforce with long prompts in a single chat, written down as lane contracts in YAML files you can read and edit.
The project is active and fairly well staffed for its age. Created in February 2026, it has about 1,836 stars, 248 forks and 25 contributors; v0.19.0 shipped on 22 June 2026 and commits were still landing on 14 October. The CLI is published to npm and crates.io, though npm counted only 119 downloads in the last week, which suggests most users run the desktop build or the source.
The cost is breadth. The README lists ACP, MCP, A2A, AG-UI, A2UI, REST and SSE as integration surfaces, two backends that must stay in step through an OpenAPI contract, worktrees, schedules, webhooks and a harness monitor. That is a platform, and a solo developer who wants parallel agents on a few branches will get there faster with Claude Squad or Vibe Kanban. Several lanes also mean several model calls per card, so token spend rises with the strictness.
The lane prompts are worth reading even if you never install it: they are a clear specification of what a coding agent should produce before anyone trusts its output.
Where it earns its place
A tech lead of a six-developer product team
Stop agent pull requests arriving without tests or with unrelated changes
Route agent work through the Kanban lanes so Review Guard checks each acceptance criterion and rejects scope creep or a dirty git state before a person looks. The software is free; each card costs several model calls.
An engineering manager at a regulated company
Keep a record of what an agent was asked, what it changed and why it was approved
Self-host the web mode with Postgres via docker compose; every card accumulates the story YAML, plan, Dev Evidence, review findings and a completion summary.
A developer exploring a new repository
Get an architecture overview before planning work
npm install -g routa-cli, then routa -p "Explain the architecture of this repository" against a local checkout; move to the board only once there is work to split up.
Before you commit
- Setup
- Download the desktop build from GitHub Releases, create a workspace, enable one provider and attach a repository; the quick start aims for a first useful answer in about five minutes. The web mode runs from source or docker compose.
- Integrations
- Coding agents and providers through ACP and adapters; MCP tools and custom MCP servers; A2A, AG-UI, REST and SSE; GitHub repositories, issues and pull requests; schedules and webhooks.
- Data and privacy
- Local-first in desktop mode; self-hosted web mode can use Postgres. MIT licence. Code and prompts go to whichever model providers you enable.
- Limits
- Web install needs npm --legacy-peer-deps; 6 open issues; much of the docs and community activity is split between English and Chinese.
What’s good
- Review is a hard gate with explicit rejection rules
- Each lane writes evidence onto the card, leaving an audit trail
- Desktop, CLI and self-hosted web from one codebase
- Lane and role prompts are plain YAML you can edit
What’s not
- Large surface area: seven protocols and two backends
- Several specialist passes per card multiply token use
- No tagged release since June 2026 despite ongoing commits
Best for
Teams that already run coding agents and want a visible, rule-based path from request to reviewed commit.
Routa alternatives
Routa: common questions
- Is Routa worth it?
- A thoughtful process for making coding agents prove their work at each step, wrapped in more surface area than most small teams will want to learn. We rate it 3.5 out of 5.
- How much does Routa cost?
- Free and open source under the MIT licence; model and agent costs come from the providers you connect.
- What are the downsides of Routa?
- Large surface area: seven protocols and two backends; several specialist passes per card multiply token use; no tagged release since June 2026 despite ongoing commits.
- Who is Routa best for?
- Teams that already run coding agents and want a visible, rule-based path from request to reviewed commit.
- Is Routa safe for company data?
- Local-first in desktop mode; self-hosted web mode can use Postgres. MIT licence. Code and prompts go to whichever model providers you enable.
- What are the best Routa alternatives?
- The closest alternatives we have reviewed are Vibe Kanban, Claude Squad, OpenHands and BMAD Method.
Answers checked Nov 3, 2026.