Dafthunk
Dafthunk is an open source visual workflow automation platform. You connect nodes in a drag-and-drop editor to build AI, scraping and data pipelines that run serverless on Cloudflare Workers, under a permissive MIT license.
What is Dafthunk?
Dafthunk is an open source visual workflow automation platform built for Cloudflare's serverless stack. Workflows are assembled in a React Flow editor: nodes are dragged onto a canvas, wired input to output, previewed live and versioned, then executed on Cloudflare Workers and Cloudflare Workflows. Execution is durable, so long-running workflows survive restarts and retries, and it scales to zero when nothing is running. State is persisted in D1 for SQL, R2 for objects, Durable Objects, KV and Workers Analytics Engine.
The home page advertises more than 470 nodes, over 30 AI models and more than a dozen integrations, organized in six families: artificial intelligence, browser automation, data processing, media generation, integrations and developer tools. AI is treated as a primitive rather than as one connector among others, with Claude Opus 4.1, Claude Sonnet 4, GPT-5, Gemini 2.5 Pro, Cloudflare and Replicate models called straight from nodes. Browser automation covers scraping, structured extraction, screenshots and PDF generation; data nodes handle JSON, CSV, text and mathematical operations; media nodes produce images, audio, documents and 3D models.
Workflows start from an HTTP request or webhook, a queue, an email, a cron schedule, a Discord or Telegram bot, or a manual run, and they draw on shared resources: typed schemas, managed SQLite databases, datasets for retrieval-augmented generation, encrypted secrets that never appear in definitions or logs, and OAuth connections to Google, Slack and GitHub. Everything lives inside an Organization, which carries roles and permissions for team work.
The licensing position is the project's central argument. The whole codebase is MIT licensed, with, in the publisher's words, no enterprise tier, no fair-code carve-outs and no community edition hiding the good parts, and the same code powers the hosted SaaS and a self-hosted deployment on the user's own Cloudflare account. The stack is published openly, React 19, TypeScript, TailwindCSS and Shadcn UI on the front end, Hono, Cloudflare Workers, Cloudflare Workflows and D1 on the back end, and development happens in public on GitHub. The publisher describes the project as an exploration of AI-assisted development and states that it sits at alpha release stage, under active development.
What it does
- Build automation workflows visually by connecting nodes in a drag-and-drop React Flow editor.
- Run AI models inside a workflow, from Claude, GPT and Gemini to Cloudflare Workers AI, for text generation, vision and natural language processing.
- Automate a browser to scrape pages, extract structured data, capture screenshots and generate PDF files.
- Generate and transform media: images, audio, documents and 3D models.
- Trigger workflows by HTTP webhook, message queue, cron schedule, email, Discord or Telegram bot, or manually.
- Expose any node as a tool that an AI agent can call.
- Invoke workflows from outside through the REST API, authenticated with a Bearer API key.
When to use Dafthunk / When not to
A quick filter to help you decide if Dafthunk is the right fit.
When to use Dafthunk
- Development teams already building on Cloudflare, whose Workers, D1, R2 and Workers AI bindings put automation on the very same infrastructure as the rest of their stack.
- Engineers who want automation that costs nothing when nothing is running: workflows scale to zero and Cloudflare is billed only on actual execution.
- Projects that need a genuinely permissive license, since the MIT source can be forked, embedded in a commercial product or resold as a hosted version.
- AI-oriented builders assembling pipelines around image generation, transcription, summarization, translation and sentiment analysis, with Claude, GPT, Gemini and Workers AI models available as native nodes.
- Data and content teams running scraping, screenshot capture or PDF generation at scale through Cloudflare Browser Rendering, on a platform that self-hosters deploy on their own account.
When not to use Dafthunk
- Organizations that require an air-gapped or fully on-premise deployment: a Cloudflare account is mandatory, and the publisher states plainly that Dafthunk does not fit air-gapped environments.
- Teams looking for the widest catalog of SaaS connectors, in CRM, marketing or project management, where n8n, Zapier and Make cover far more services than the six integration providers named here.
- Regulated or risk-averse teams handling critical data, since the publisher itself advises against processing highly sensitive or critical information through the service at this alpha stage.
- Buyers who need contractual assurance before adoption: no paid SLA, no dedicated support channel, no published data processing agreement and no security certification are documented.
- Anyone who needs a predictable hosted budget, because no pricing page exists even though the legal documents confirm a paid subscription processed by Stripe.
How to use Dafthunk
A typical end-to-end flow, from setup to results.
- Open app.dafthunk.com and sign in with Google, GitHub or an email account.
- Work inside an Organization, the container that holds workflows, triggers, resources and executions, with its own roles and permissions.
- Start from one of the ten published templates, from image generation to speech to text, screenshots, summarization or translation, or open a blank canvas.
- Drag nodes onto the React Flow canvas and connect their inputs and outputs to define the data flow and the logic.
- Configure the shared resources the workflow needs: typed schemas, managed SQLite databases, RAG datasets, encrypted secrets and OAuth integrations.
- Attach a trigger, whether an HTTP request or webhook, a queue, an email, a cron schedule or a Discord or Telegram bot, then enable the workflow, which will not run otherwise.
- Run it and follow the execution, whose status (submitted, executing, completed, error), inputs, outputs and logs are recorded for monitoring and debugging.
- Iterate on the versioned workflow; the publisher claims a first workflow takes about four minutes and the hundredth about forty seconds.
- To call a workflow from your own code, generate an API key in Settings then API Keys, keeping in mind it is displayed only once, and send it as an Authorization Bearer header to api.dafthunk.com.
- To self-host, clone the dafthunk-com/dafthunk repository, run pnpm install and pnpm dev, then deploy to your own Cloudflare account with Workers, Workflows, D1, R2, Workers AI and Analytics Engine.
Pros & Cons
Pros
- Unrestricted MIT license: forking, commercial embedding and reselling a hosted version are all permitted, with no enterprise tier, no fair-code carve-out and no stripped-down community edition.
- No idle cost, since workflows scale to zero and Cloudflare is billed on actual execution only.
- Nothing to provision: no container, no external database and no always-on virtual server.
- Durability built into the runtime, so long-running workflows survive restarts and retries.
- AI handled as a primitive rather than as an integration category, with native bindings to Workers AI, OpenAI, Anthropic and Gemini.
- Reversibility between the hosted SaaS and a self-hosted deployment, both running the same code, so a team can switch either way at any time.
- Development in the open on GitHub with contributions welcome, a REST API documented in three languages and ten ready-made workflow templates.
Cons
- Declared alpha release under active development, with no guarantee of availability, accuracy, completeness or suitability.
- Complete dependence on Cloudflare: an account is mandatory, and no air-gapped deployment is possible.
- No public pricing page, although the privacy and cookie policies both confirm paid subscriptions and a Stripe checkout.
- Thin privacy governance: no data processing agreement, no data protection officer, no Article 27 EU representative, and not a single occurrence of the word GDPR.
- No corporate identity published, with no postal address and no full legal name beyond the mention Location: Switzerland.
- No security certification such as SOC 2 or ISO 27001, no named data hosting country, and international transfers announced without a detailed mechanism.
- A far narrower SaaS integration library than n8n or Zapier, a web-only product with no mobile app or browser extension, and support routed to GitHub issues rather than a dedicated channel.
Pricing & Plans
A permanently free route exists. The software is published in full under the MIT license and can be self-hosted at no license cost, the user paying only actual Cloudflare consumption, which the publisher estimates at a few cents per thousand executions. The hosted service is presented with a Start for free call to action, and the structured data on the home page declares an offer priced at 0 USD. A paid subscription nevertheless exists: the privacy policy refers to subscribing to a paid plan and to payment processing by Stripe, and the cookie policy documents a checkout redirection to the payment provider. No pricing page appears on the site or in its 103-URL sitemap, and no amount, currency or billing unit is published, so no lowest paid price point can be stated.
- Hosted SaaS
- entered through a Start for free call to action on app.dafthunk.com
- no tier name
- scope or price is published.
- Self-hosted deployment of the full MIT source on the user's own Cloudflare account
- free of license cost and billed only by Cloudflare on actual usage.
- Paid subscription
- whose existence is confirmed by the privacy policy and by the Stripe checkout described in the cookie policy
- but whose name
- feature scope and price are not published.
Data, GDPR & hosting
A consolidated view of how Dafthunk handles your data.
GDPR overview
The word GDPR appears nowhere on the site, and no compliance is claimed. The privacy policy, effective 9 June 2025, nevertheless applies the regulation's machinery: a Legal Basis for Processing section citing contractual necessity, legal obligation, legitimate interests and consent, and a Your Rights section granting access, correction, erasure, objection or restriction of processing, and portability, all exercised at contact@dafthunk.com and qualified by the phrase Depending on your location. A consent banner gates Google Analytics, and the minimum age is 13, raised to 16 in the European Union. The gaps are equally clear: international transfers are announced, since the service is accessible worldwide, without standard contractual clauses or any adequacy mechanism; no data protection officer is named; no Article 27 EU representative is designated; and no data processing agreement is published.
Who owns the data?
Neither the terms nor the privacy policy, both effective 9 June 2025, claim ownership of user content: the publisher positions itself as the operator of a service that processes what users push through their workflows. It stores account data (username, email address, and the basic Google or GitHub profile when social sign-in is used), workflow definitions, inputs, outputs, intermediate results and execution metadata, while billing data is handled by Stripe. Data drawn from third-party integrations, including Gmail and Google Calendar, is accessed only during workflow execution but may persist in execution logs. Users remain solely responsible for holding the rights and permissions attached to the data they process. Self-hosters, deploying on their own Cloudflare account, keep full control of data, deployment and costs.
Reuse rights
The privacy policy states that personal information serves to provide, maintain and improve the service, process subscriptions and payments, communicate about the account and answer support requests; usage logs and interaction data feed product improvement. The declared legal bases are contractual necessity, legal obligation, legitimate interests and consent. Sharing is limited to Stripe for payment processing and Cloudflare for hosting and content delivery, with no other third party except where the law requires it, while Google Analytics loads only after consent given through an accept or reject banner. The publisher explicitly rules out advertising uses and the sale of personal information. Training AI models on customer data is never addressed in the documents read, so no commitment exists either way. The terms read likewise contain no clause on the ownership or reuse of generated outputs: no explicit permission and no explicit restriction is published, and reuse therefore remains governed by the rights attached to the underlying data and by the policies of the AI providers called from the nodes.
Data retention & training
Hosting summary
Hosting rests entirely on Cloudflare. The privacy policy names Cloudflare as the provider for hosting and content delivery, and the documentation confirms that workflows execute on Cloudflare Workers and Cloudflare Workflows, with state persisted in D1 SQL databases, R2 object storage, Durable Objects, KV and Workers Analytics Engine. No hosting country and no region is named anywhere in the pages read, and no residency option is offered. The policy, effective 9 June 2025, states that because the service is accessible worldwide, personal information may be transferred and stored outside the user's country of residence, and it describes no standard contractual clauses or adequacy mechanism for those transfers. Network checks resolve the domain to 188.114.96.12 on AS13335 (Cloudflare), an anycast node nominally geolocated in the United States, which is a content delivery point of presence rather than evidence of where data is stored. The publisher declares Location: Switzerland in its legal pages, without a postal address. Self-hosting changes the picture entirely, since the deployment runs on the user's own Cloudflare account and the account holder chooses the configuration.
Things to keep in mind
Risks and trade-offs to weigh before adopting Dafthunk.
- Alpha software: the terms warn of bugs, errors and unexpected behaviors, disclaim every guarantee and place the risk squarely on the user.
- Parts of the service, including core functionality, are developed using generative AI technologies, which the terms themselves flag as an additional risk factor.
- The publisher advises against processing highly sensitive or critical information at this stage and recommends backing up important data regularly.
- Data pulled from third-party integrations such as Gmail, Google Calendar or Slack transits through the platform and may be retained in execution logs, so an over-permissive OAuth scope exposes far more than a workflow strictly needs.
- Workflows fired by a schedule, a webhook or a bot run without a human in the loop: an unreviewed AI output, a wrong summary, a hallucinated value or an automated message can propagate at scale before anyone notices, and habit quickly turns unchecked automation into unquestioned truth.
- Account data may be transferred in the event of a merger, acquisition, financing, bankruptcy or asset sale, and no data processing agreement, EU representative or security certification is published to frame that risk.
- Infrastructure lock-in and unpredictable hosted cost: the product runs only on Cloudflare, and the paid subscription referenced in the legal documents comes with no published price grid.
Setup & Integrations
Technical difficulty
Two levels. The hosted service is straightforward: sign up on app.dafthunk.com, sign in with Google or GitHub, no infrastructure prerequisite, ten ready-made templates, and a claimed first workflow in about four minutes, the visual editor being designed to keep users away from the command line. Calling workflows from your own code adds a light developer step, namely generating an API key and sending a Bearer header, with cURL, JavaScript and Python examples provided. Self-hosting is genuinely technical: clone the repository, run pnpm install and pnpm dev, then deploy to a Cloudflare account.
Deployment
Integrations
Behind Dafthunk
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Dafthunk.
Frequently asked questions
What is Dafthunk?
Is Dafthunk free and open source?
How does Dafthunk differ from n8n, Zapier or Make?
Do I have to manage servers or containers?
What kind of workflows can be built?
Can Dafthunk be self-hosted on my own Cloudflare account?
Is there an API?
How mature is the product?
Who processes the data?
What is the minimum age to use Dafthunk?
Should you pick Dafthunk?
Dafthunk is a deliberately narrow proposition: visual workflow automation for people who accept living inside Cloudflare's ecosystem. Within that boundary the offer is coherent. The platform is MIT licensed end to end, with no enterprise tier and no fair-code restriction, which sets it apart from n8n and gives teams the right to fork it, embed it in a commercial product or run a hosted version of their own. Execution on Cloudflare Workers and Workflows removes the servers, containers and always-on databases that self-hosted automation usually demands, and idle time costs nothing. More than 470 nodes, with AI models, browser automation and data processing treated as first-class citizens, cover a real range of use cases, and a documented REST API opens those workflows to the rest of the stack.
The reservations are just as clear, and most of them come from the publisher itself. The terms describe an alpha release under active development, with no guarantee of availability or accuracy and parts of the service built with generative AI. The privacy policy, effective 9 June 2025, advises against processing highly sensitive information at this stage; it applies GDPR mechanisms without ever naming the regulation, publishes no data processing agreement, no EU representative and no security certification, and names no hosting country. No pricing page exists, even though three legal documents confirm a paid subscription processed by Stripe, so the cost of the hosted route cannot be anticipated. Corporate identity stops at a brand name and the mention Location: Switzerland.
The natural audience is technical: developers, data and AI profiles, freelancers and small teams already comfortable with Cloudflare, who value a permissive license and reversibility more than contractual guarantees. Organizations bound by procurement, compliance or air-gap requirements will want to wait until the project leaves alpha and documents its governance.
- Choosing a selection results in a full page refresh.
- Opens in a new window.