Coordiñareby Vivi Dynamics

Your agents, working in concert.

Bring your coding agents together. Coordiñare takes work from your GitHub board through planning, implementation, and review. You hold the final approval.

Get started View on GitHub
  • Isolated execution environments
  • Flexible agent & model backends
  • GitHub-native workflows
The Coordiñare conductor, a smiling silver-haired gentleman in a green tailcoat, guiding with his baton.

One workflow. Every part in place.

AssessCheck the card has context
ArchitectCommit a plan
ImplementWrite code, open a PR
ReviewApprove or bounce
VerifySecurity and QA
MergeA human approves

Failures route back to the implementer, then the affected stages re-run.

From issue to implementation.

Your team reviews finished changes. Coordiñare handles everything between the card and the pull request.

  1. 01

    Understand

    Coordiñare picks up a card and checks it has enough context to act on.

  2. 02

    Plan

    An architect agent reads the codebase and commits a technical plan to the branch.

  3. 03

    Build

    An implementer writes the code in an isolated environment and opens a pull request.

  4. 04

    Verify

    Review, security, and QA agents check the work. Failures go back to the implementer.

Specialized agents. One coordinated workflow.

Every card moves through a sequence of roles. Configure the ones you want; the rest are skipped.

01Prepare
Advocate

Adds high-value issues to the board.

Assessor

Checks a card has enough context to start.

Architect

Reads the codebase and commits a plan.

02Build
Implementer

Writes the code and opens the pull request.

Feedback from any later stage comes back here.

03Verify
Reviewer

Reviews the PR. Approves or requests changes.

Security

Scans the change for vulnerabilities.

QA

Validates acceptance criteria and writes tests.

04Finish
Tech Writer

Updates the changelog, README, and docstrings.

Closer

Confirms feedback was addressed and resolves threads.

Built to fit the way you already work.

Coordiñare works with the stack, models, and board you already have.

Agent orchestration

Each card runs through a sequence of specialized roles, with gates between them.

Any harness, any model

Use supported coding harnesses with hosted or self-hosted models.

claude_codecodex opencodejunie pihermes openclaw

Isolated execution

Agents work in containers with your real toolchain and services.

Many projects at once

Run multiple project boards while controlling how much work happens in parallel.

Per-role models and personas

Point each role at a different model, and give it standing instructions.

Visibility and alerts

A live dashboard, per-card cost tracking, and Slack or email alerts.

Real-time visibility

See the work as it happens.

The dashboard streams live updates as cards move, so you can see what's running and what's stuck.

  • Live workflow status and current performer
  • Project board and card tracking
  • Token and cost usage per card
  • Health checks and activity feed
Project workflowExample board

Implement

Add search to the help centerWriting code

Implementer working

Review

Improve keyboard navigationIn review

Reviewer checking changes

Approval

Fix the export date formatChecks passed

Waiting for human approval

An illustrative workflow. Your project, its agents, and the work between them.

Autonomous, but not unsupervised.

Everything up to the final approval is automated. Several gates stand between an agent and your main branch.

  • CI has to be greenRequired checks pass before a card advances.
  • Security has to clear itThe security stage has to pass.
  • Acceptance criteria get validatedQA checks the card's criteria and writes tests.
  • Retries are boundedAfter a set number of feedback cycles, the card is blocked for a human.
  • Infrastructure failuresEnvironment problems pause the work and notify operators.

Built for teams exploring agent-driven development.

Engineering teams

Work already lives on a GitHub board. Coordiñare picks it up there.

Technical startups

Ship with AI agents without building the orchestration layer yourself.

AI developers & researchers

Experiment with harnesses, models, and lifecycle configurations on real repos.

Get started

Put your agents to work.

Python 3.12+, uv, and a GitHub project board. Configuration details are in the repo.

$ git clone https://github.com/ViviDynamics/coordinare.git
$ cd coordinare
$ uv sync
$ cp config.example.yaml config.yaml
$ .venv/bin/python -m coordinare config validate
$ bin/run-coordinare
# dashboard → http://localhost:8090
Documentation View on GitHub