Your n8n workflows on one small binary.
Flowwolf runs the workflows you built in n8n with the same JSON, the same expressions and the same API, in a 105 MB image that idles at about 5 MB of memory. Open source, written in Rust.
docker run -p 5678:5678 -v flowwolf:/data flowwolfBring the workflows you already have.
Flowwolf reads the n8n workflow format as it is. Export from n8n, paste onto the canvas, activate.
- Same nodes and parameters
- Node types, settings, connections and pinned data are read without conversion. Unknown nodes are kept and reported when they run.
- Same expressions
{{ $json.email }},$('Node'),$input,$nowand$varsrun in an embedded JavaScript sandbox.- Same API
- Scripts that call
/api/v1with anX-N8N-API-KEYheader keep working.
# import a workflow exported from n8n curl localhost:5678/api/v1/workflows \ -H "X-N8N-API-KEY: $KEY" \ -H "Content-Type: application/json" \ -d @new-lead.json # and switch it on curl -X POST \ localhost:5678/api/v1/workflows/$ID/activate \ -H "X-N8N-API-KEY: $KEY"
Everything a workflow tool needs, in one process.
Editor, engine, scheduler and webhooks run together. Back up one folder or point it at PostgreSQL.
Visual editor
Map fields from the input panel, pin test data and run single steps until the flow does what you want.
JavaScript and Python
The Code node runs JavaScript in QuickJS and Python in a separate runner process the kernel keeps away from files and network.
Triggers
Webhooks, forms, chat, schedules, mail, MQTT, Redis, RSS and file changes. Over forty service triggers register their webhook and verify signed events, and polling triggers watch Gmail, Sheets, Notion and more.
AI agents
n8n-style AI agents and chains with sub-nodes: twelve model providers, memory in Postgres or Redis, tools, MCP client and server, RAG with Qdrant, Pinecone, Supabase or PGVector.
Teams
Owners, admins and members. Share workflows and credentials as viewer or editor. Credentials are encrypted, OAuth2 connects in the browser.
Executions you can read
Every run keeps the data of each node. Open it on the canvas, find the failing step and retry. Waiting runs survive a restart.
Source control
Push and pull workflows to any Git server over SSH with a deploy key Flowwolf generates, or through the GitHub and Gitea APIs.
Scale out
Queue mode hands production runs to worker processes. With PostgreSQL the workers can run on other machines.
Runs on a Raspberry Pi. Scales to a cluster.
One container and one volume is a complete installation. Add workers and PostgreSQL when the load grows.
# one machine docker run -d --name flowwolf -p 5678:5678 \ -v flowwolf:/data \ -e GENERIC_TIMEZONE=Europe/Berlin flowwolf # several machines: main plus workers on PostgreSQL -e DB_TYPE=postgresdb -e DB_POSTGRESDB_HOST=db \ -e EXECUTIONS_MODE=queue -e FLOWWOLF_ENCRYPTION_KEY=… docker run … flowwolf worker
- No runtime to manage
- No Node.js, no npm, no Redis. The binary contains the editor, the engine and every node.
- SQLite or PostgreSQL
- SQLite in the data folder by default. PostgreSQL with the same variables n8n uses.
- Workers that never double up
- Jobs are claimed with row locks, so two workers never run the same execution.
408 node types, from Slack to AWS.
Each one is a real node with operations and stored credentials. A selection:
Communication
Slack, Microsoft Teams, Discord, Telegram, WhatsApp, Mattermost, Matrix, Gmail, Outlook, SMTP and IMAP, Twilio, Postmark, SendGrid, Mailjet
Data and storage
PostgreSQL, MySQL, Redis, MQTT, Google Sheets, Excel 365, Airtable, Notion, Supabase, NocoDB, DynamoDB, Firestore, Google Cloud Storage
Sales and support
HubSpot, Salesforce, Pipedrive, Zoho CRM, Shopify, Shopware 6, WooCommerce, Stripe, Zendesk, Freshdesk, Intercom, Help Scout
Developer and operations
GitHub, GitLab, Bitbucket, Jira, Linear, Sentry, PagerDuty, Grafana, Jenkins, CircleCI, Cloudflare, Netlify, Okta
AI
AI Agent and chains with sub-nodes, twelve chat model providers, MCP client and server, vector stores in Qdrant, Pinecone, Supabase and PGVector
Cloud platforms
AWS Lambda, SQS, SNS, SES, Textract and more with Signature V4, Google Workspace, Microsoft Graph, Entra ID, SharePoint
Anything else is one HTTP Request node away, with stored credentials and OAuth2.
Compared with n8n.
Flowwolf follows the n8n format so you can move in both directions. Where they differ:
| Aspect | Flowwolf | n8n |
|---|---|---|
| Runtime | One Rust binary | Node.js with npm packages |
| Docker image | ~105 MB | Several hundred MB |
| Database | SQLite or PostgreSQL | SQLite or PostgreSQL |
| Queue mode | Workers on PostgreSQL, no Redis | Workers on Redis and PostgreSQL |
| Workflow format and API | n8n JSON, /api/v1 | n8n JSON, /api/v1 |
| Integrations | 408 node types | More than 400 |
| License | MIT | Sustainable Use License |
Questions people ask first.
Can I import my existing n8n workflows?
Yes. Export a workflow in n8n and paste the JSON onto the canvas, or send it to the API. Nodes Flowwolf does not know yet stay in the workflow and report an error when they run.
How do expressions work without Node.js?
Flowwolf embeds QuickJS, a small JavaScript engine. Each run gets its own sandbox with $json, $input, $('Node'), $now and $vars. Endless loops stop at a time limit.
Where is my data stored?
In one SQLite file in the data volume, next to the key that encrypts credentials, or in PostgreSQL if you set DB_TYPE=postgresdb. Old executions are pruned automatically.
Is it safe to put on the internet?
Passwords are hashed with Argon2, sessions use HttpOnly cookies, credentials are encrypted, file access is limited to one folder, Python runs in an isolated process and the shell command node is off by default. Run it behind a TLS reverse proxy like any web app.
What does it cost?
Nothing. Flowwolf is open source under the MIT license, for yourself, your team or your customers.
Start in a minute.
Run the container, create the owner account and paste in your first workflow.
docker run -p 5678:5678 -v flowwolf:/data flowwolf