Skip to main content
Flowfield is a local service that coordinates AI coding agents across parallel tasks, keeping every piece of a task’s lifecycle — intent, activity, questions, answers, and results — together in one shared feed. You interact with it through a browser UI at localhost:8765 and a CLI, while your AI coordinator plans work and workers implement agreed tasks in isolated Git checkouts. Nothing leaves your machine: Flowfield stores all data in ~/.flowfield/ and requires no cloud account.

Introduction

Learn what Flowfield is, how it works, and who it’s for.

Quick Start

Install Flowfield, connect Codex, and run your first task in one walkthrough.

Core Concepts

Understand tasks, feeds, the board, workers, and the review workflow.

Codex Integration

Connect your Codex coordinator to Flowfield via MCP.

​
Get up and running

1

Install Flowfield

Install the flowfield-core package using uv with Python 3.12:
The package bundles the CLI, service, and browser UI — no separate frontend install is needed.
2

Start the service

Start the local service and open the browser UI:
Open localhost:8765 in your browser. Keep the terminal open; Ctrl-C stops the service.
3

Initialize your project

In a second terminal, change to your project directory and run the setup commands:
Review the generated files (.flowfield/config.toml and .agents/skills/flowfield-coordinator/SKILL.md) and commit them before running workers.
4

Start a Codex conversation and run your first task

Open a fresh Codex conversation in your project directory. Describe a small change, ask your coordinator to capture the task in Flowfield, move it to Up Next, and click Run Queue. Monitor progress in the browser, answer any questions in the task feed, and approve the result when it reaches In Review.

​
Why Flowfield

Shared Task Feeds

Intent, worker activity, questions, answers, and results live in one feed per task — nothing gets lost across tools or conversations.

Parallel Workers

Independent tasks run simultaneously in isolated Git checkouts. Dependent tasks wait automatically until prerequisites are complete.

Explicit Review

No code reaches your project without your approval. Inspect the diff, try the result in an independent copy, request changes, then approve and integrate.