NoClick

Describe an agent. Watch it get built. Let it run.
Open-source workflow automation with AI agents at the core — a visual canvas, ~160 integrations, and coding agents that can actually use your tools.

Website · Docs · Self-hosting · Discord · Contributing

Quick start

One line:

# macOS / Linux / WSL2 -- needs Docker: https://docs.docker.com/get-docker/
curl -fsSL https://noclick.com/install.sh | sh

# Windows PowerShell -- installs Docker Desktop itself; Windows restarts once
# the first time, and the installer continues after you sign back in
irm https://noclick.com/install.ps1 | iex

Already have Docker, Node.js 18+ and Git? The launcher the Windows installer hands off to runs everywhere:

npx noclick

Or by hand — the same thing, visibly:

git clone --branch <tag> https://github.com/noclickapp/noclick.git && cd noclick
./scripts/noclick-setup.sh
NOCLICK_VERSION=<version> docker compose up -d

Each installer path fetches the latest release's source, generates this instance's secrets — including the credential-encryption key, which is kept across re-runs — and pulls that release's images: about 1.5 GB to download the first time, and nothing is compiled on your machine. Re-running updates in place. Pin a release with NOCLICK_REF=<tag> from the releases page; NOCLICK_REF=main runs the development line. The frontend image is built for the default local URLs, so an install with other public URLs builds that one image itself, and NOCLICK_BUILD=1 builds everything from the checkout. Every installer option is documented at the top of install.sh and install.ps1.

Then open the editor at http://localhost:3000.

What is NoClick?

NoClick lets you build automations on a visual canvas and run them on triggers — webhooks, schedules, inbound email, or chat messages from Slack, WhatsApp and friends. What makes it different is the agent layer: an AI agent node can drive real coding harnesses (Claude Code, Codex, opencode, hermes, OpenClaw) and call your connected integrations as tools, so an agent can read a Linear issue, work in a git repo, and post the result back to Slack.

This repository is the whole platform, open source under the AGPL: the workflow engine, every integration node, the React editor, the realtime collaboration layer, the AI workflow builder, and an MCP server. It's the same code that runs noclick.com, and it runs end to end on your own machine with nothing held back.

What you can build

• Automations on a canvas — chain ~160 integrations (Slack, Gmail, Notion, Linear, HubSpot, Stripe, Postgres, Google Sheets, …) with branching, iteration, filters, and code nodes.
• Triggers — webhooks, cron schedules, inbound email, and app events from Slack/Discord/HubSpot; channel-style agents that reply where they're spoken to, with as many triggers per workflow as you need.
• AI agents with real tools — wire integration nodes into an agent's tool handle and it can call those operations directly. Agent nodes run either in-process (any OpenAI/Anthropic/OpenRouter-compatible model) or as a coding harness using your own installed CLI.
• Interfaces — publish a form, dashboard, or chat over any workflow.
• Build with AI — describe what you want and the builder assembles the workflow on the canvas. It's also exposed over MCP, so you can point Claude Code (or any MCP client) at your instance and have it build for you.

How it works

1. Describe it. Type what you want into the builder — or point any MCP client at your instance and let your coding agent do the describing.
2. The builder assembles it. Nodes appear on the canvas wired to real operations, asking only for what it genuinely needs: credentials, and the choices only you can make.
3. It runs on triggers. A message arrives, a webhook fires, a schedule ticks — the workflow executes, agents call their tools, and results land where you told them to. Every run records its tool calls, timings, and the agent's response, so you can see exactly what happened.

Self-hosting

The installer above is the fast path. For a real deployment — your own Postgres, object storage, OAuth apps for integrations, model providers, and the environment variables each side needs — read the self-hosting guide.

The deploy buttons above ask for nothing: Railway and Render create a Postgres database next to the reviewed release image (DigitalOcean binds a managed cluster named noclick-db you create first), and the instance does the rest on first boot — it runs its own auth layer, prepares the database, and generates its own keys. Railway uses railway.template.json, Render render.yaml, DigitalOcean .do/deploy.template.yaml, and Fly fly.toml. See the hosted deployment guide for what each one provisions and where its secrets live.

Releases are what you run: main moves with every upstream merge, tags are deliberate cuts, and each tag publishes images to GHCR so the compose stack pulls rather than builds. Pin with NOCLICK_VERSION.

Developing from source

Prerequisites: Docker, the Supabase CLI, Python 3.12 (3.13 isn't supported yet — some pinned wheels don't build on it), Node.js 20 or newer, and pnpm 10.34.4 (or npm for running package scripts).

pip install -r requirements.txt
make local

That boots everything on your machine — a local Supabase (Postgres, auth, storage), the backend, and the frontend — and prints a URL. Sign up with any email; local auth confirms instantly. No cloud account, no API keys required to get to a running canvas.

# Backend
cd backend && python server.py       # or: uvicorn server:web_app
pytest                               # test suite

# Frontend
cd frontend && npm run dev
npm run typecheck && npm run build && npm test

# Database migrations
cd infra && supabase migration up --local

Architecture

Node config UIs are generated from the backend's Pydantic models, so adding an integration is mostly writing one Python module — the editor picks it up.

NoClick Cloud

noclick.com is the hosted platform. It runs this engine plus managed infrastructure: retained agent runtimes so turns resume instantly, scaled webhook and cron delivery, additional hosted builder capabilities, and managed storage and email. Self-hosted and cloud share the same workflow format, so workflows move between them unchanged, and the cloud's operational layer is what funds the open-source engine.

Contributing