flow

# Aixle Flow **The team-level control plane for AI coding agents.** Put a card on the board, an agent picks it up in an isolated container, your team sees the run, the cost, and the output — without anyone opening a terminal. [](LICENSE) [](https://rubyonrails.org/) [](https://react.dev/) [](ROADMAP.md) [](https://github.com/AixleHQ/flow/actions/workflows/ci.yml?query=branch%3Adevelop) [](https://github.com/AixleHQ/flow/actions/workflows/ci.yml?query=branch%3Adevelop) [](https://github.com/AixleHQ/flow/actions/workflows/ci.yml?query=branch%3Adevelop)  ## Why Aixle Flow Cursor and Claude Code are personal superpowers. They run on **your** laptop, in **your** terminal, for **your** task. Once more than one person on a team starts delegating work to agents, the wheels come off: - Nobody knows what's running, where, or whether it's stuck. - There's no shared trail of what an agent did or how much it cost. - Setup is per-developer, so the team uses agents five different ways. Aixle Flow is the layer above the agent — the Kanban board, the workflow graph, the audit log, the cost meter. The agent stays great at coding; Aixle Flow makes it usable by a team. It is the only open platform we know of that combines: - A **board** with column → workflow bindings (drop a card, trigger a run) - A **DAG workflow engine** with retries, approval gates, and parallel steps - **Container-isolated** agent execution (Claude Code, Cursor CLI, Codex, Gemini CLI, Antigravity CLI, Grok CLI, Kiro CLI) - **MCP-native** tool integration - **Per-session cost tracking** in tokens and cents - **Temporal** under the hood for durable, long-running runs ## Quickstart Three commands to a working app on `http://localhost:4000`. ```bash git clone https://github.com/AixleHQ/flow.git && cd flow cp .env.example .env.development # fill in OAuth secrets if you want SSO make setup # builds containers, installs deps, seeds DB ``` Then start everything in one terminal: ```bash make up # web, worker, db, redis, traefik, temporal ``` Open `http://localhost:4000` and sign in. > **5-minute setup is on the roadmap.** Right now `make setup` builds > agent images which takes longer the first time. See > [`ROADMAP.md`](ROADMAP.md) for the planned `docker compose up` > single-command flow. Need more detail? See [docs/quickstart.md](docs/quickstart.md). ## Documentation Three tiers, each independently navigable: | Tier | For | Where | | ------------ | ------------------------------ | ------------------------------------------- | | **Quickstart** | First-time users, < 5 min | [docs/quickstart.md](docs/quickstart.md) | | **User Guide** | Common workflows & gotchas | [docs/user-guide/](docs/user-guide/) | | **Reference** | Full API / CLI / config surface | [docs/reference/](docs/reference/) | ## Roadmap What's planned and what's in flight — see [ROADMAP.md](ROADMAP.md). Issues tagged [`good first issue`][gfi] are a good place to start contributing. [gfi]: https://github.com/AixleHQ/flow/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22 ## Tech Stack - **Backend** — Ruby on Rails 8, PostgreSQL, Redis, Temporal - **Frontend** — React 19, Inertia.js, Mantine 9, Vite, TypeScript - **Agents** — Claude Code, Cursor CLI, OpenAI Codex CLI, Gemini CLI, Antigravity CLI, xAI Grok CLI, Kiro CLI (each runs in its own container — Docker locally, Kubernetes in production) - **Protocols** — MCP for tools, ActionCable for live updates - **Dev environment** — Docker Compose, one Makefile ## Inspired by Aixle Flow stands on the shoulders of the open agent-and-workflow ecosystem. A few projects shaped how we think: - **[n8n](https://n8n.io)** — workflow automation can be open, code-friendly, and priced per run rather than per step. - **[Dify](https://dify.ai)** and **[Langflow](https://www.langflow.org)** — visual, open-source orchestration of LLM agents and pipelines. - **Paperclip** — treating a roster of agents like an org, with budgets, governance, and per-agent cost as first-class concerns. - **[Cursor](https://www.cursor.com)** — background agents that run in isolation and are triggered by events, not just a human at a keyboard. - **[Temporal](https://temporal.io)** — durable execution as the right foundation for long-running, retryable agent runs. - **[Model Context Protocol](https://modelcontextprotocol.io)** — an open standard for giving agents tools. - The **[Kanban board](https://en.wikipedia.org/wiki/Kanban_board)** — work lives in columns anyone on the team can move. Where each inspired a category, Aixle Flow's own contribution is to make it **board-native and team-first**: drop a card, an agent runs in a container, and the whole team sees the run, the trail, and the cost. ## Contributing We welcome PRs, issues, and discussion. 1. Read [CONTRIBUTING.md](CONTRIBUTING.md). 2. Pick up a [`good first issue`][gfi] or open a discussion before starting on something larger. 3. Run `docker compose exec -T web make check_all` before opening a PR. ## License Aixle Flow is released under the [Apache License 2.0](LICENSE) — a permissive license that lets you use, modify, self-host, and commercially deploy it freely, including in closed-source products, as long as you preserve the license and attribution notices. See [`LICENSE`](LICENSE) for the full terms, [`NOTICE`](NOTICE) for attribution, and [`THIRD-PARTY-LICENSES.md`](THIRD-PARTY-LICENSES.md) for the licenses of bundled dependencies. "Aixle" and "Aixle Flow" are trademarks — see [`TRADEMARK.md`](TRADEMARK.md). Contributions are accepted under the Apache 2.0 license and require both a Developer Certificate of Origin (DCO) sign-off and a signed Contributor License Agreement (CLA) — see [`CONTRIBUTING.md`](CONTRIBUTING.md).