Anyone can upload a prompt
Most marketplaces have zero quality gates. A listing is a title, a description, and a promise. You have no way to know whether the prompts are well-engineered, quietly broken, or actively malicious.
AI pipelines that run on your computer, inside your chat.
DeeWaanAI is a free, open-source MCP server you run locally. Build multi-step AI workflows from Claude, ChatGPT, Kimi, or any MCP client — with your own provider keys, so nothing ever runs on our servers. Share pipelines on the marketplace with signed proof that they work.
Connects to the tools you already use
ClaudeChatGPTKimiCursorAny MCP client
DeeWaanAI runs as a local MCP server on your computer. Connect any client you already use — per person, per machine.
Most marketplaces sell promises.
DeeWaanAI was built because the status quo is broken.
Your data goes who-knows-where
Most platforms run your work on their servers: your inputs, your outputs, and your API keys pass through opaque middlemen. DeeWaanAI runs only on your computer — execution never touches our servers.
Locked into one chat app
Force you to abandon the tools you already use and learn a new interface. Handing a colleague a working pipeline means everyone switching to the same app first.
How it works
Install locally
One setup command installs DeeWaanAI on your computer — no database server to manage, no cloud account needed to run pipelines.
Connect your chat
Paste your local MCP URL into Claude, ChatGPT, Kimi, Cursor, or any MCP-compatible client. Your chat becomes the interface.
Create, fork, run
Build pipelines in chat or with the visual designer, or fork one from the marketplace. Every run goes to your providers under your keys — we never execute anything for you.
For creators
Build and share certified pipelines
- IF/WHILE loops and quality gates for sophisticated workflows
- AI Auto-Designer turns plain language into optimal pipelines
- Judges never see your prompts during certification — only the published inputs and outputs
- A signed evidence record travels with your listing, verifiable offline
For users
Pipelines that work, in the chat you already have
- Every listing is published with the full test record, not a badge
- Runs on your own computer with your own API keys — your data never touches our servers
- One-line setup in any MCP client — Claude, ChatGPT, Kimi, Cursor
- Fork a marketplace pipeline and it is yours: a local copy you can read, edit, and run
Free. Open source. Yours.
DeeWaanAI is free and open source (AGPL-3.0). It runs on your computer, on your own provider keys — we never charge, and we never execute your pipelines. The hosted marketplace stores only listings and their evidence.
DeeWaanAI
Free & open source$0 forever — AGPL-3.0 open source
Everything included — create, run, certify, share.
- Custom pipeline designer with IF/WHILE loops and quality gates
- AI Auto-Designer — describe the goal, get a built pipeline
- Up to 3 of your own LLM providers in one pipeline
- Human-in-the-Loop approval gates
- Share pipelines on the marketplace with signed evidence
- Run it all from Claude, ChatGPT, Kimi, Cursor or any MCP client
- Bring your own keys: your provider bills you, we never charge per run
- Runs entirely on your computer — nothing is executed on our servers
Your only cost is your own LLM provider usage, billed by your provider directly. There is no paid tier and none is planned.
Free and open source. The hosted marketplace never runs your pipelines — it stores listings and evidence records only.
Power tools, included
Available behind feature flags while they mature — same price as everything else: free.
Batch runs (Research Mode)
Run up to 100 inputs through a pipeline in one go. Live progress, pause/resume on provider outages, and one-click CSV export.
Pipeline chaining
Feed one pipeline's output straight into another. Generation → evaluation → review becomes a single submission, with depth and cycle guards so chains never run away.
Pipeline sharing
Lend one pipeline to one named account, from your chat. By default they can run it and keep the answer but never read your prompts — and their run bills to their own keys. Revoke it any time.
Proof you can read, not a badge.
Anyone can upload a prompt. DeeWaanAI is built for the ones worth inspecting.
1
Deterministic security scan
Code, not AI. Catches injection patterns, hidden Unicode tricks, and data-exfiltration attempts. Ungameable.
2
Creator-funded test runs
Evaluation runs on the creator's own provider keys — the platform pays for zero model calls, so testing can never become a hidden cost passed back to you.
3
Published evidence
The creator's own judging panel — three judges from three different model developers, none of them a model that generated the answers — under three prompt framings. Every prompt, every score and every failure is published for you to read.
Fail-closed by design: an incomplete plan, a self-graded panel or a leaked secret leaves the listing unpublished.
Frequently asked questions
What is DeeWaanAI?
DeeWaanAI is a marketplace for AI pipelines — multi-step AI workflows you can create, share, and fork. Before a listing can be forked, its evidence has to pass structural checks: a completed test recorded by the platform, a judging panel that could not have graded its own work, a scan of the disclosed text for secrets, and the author's consent to the exact preview that went public.
What is a certified AI pipeline?
A published pipeline has passed structural checks, not an opinion poll: its test plan completed and recorded by the platform, three judges from three different model developers with none of them a model that generated the answers, every attempt disclosed including the failures, and the seller consenting to the exact public preview. What you get is the record itself — inputs, outputs, the judging prompts in full, every score — so you can judge the evidence rather than a badge.
How do I use DeeWaanAI?
Install DeeWaanAI on your computer (one setup command), then paste your local MCP URL into any MCP-compatible client — Claude, ChatGPT, Kimi, Cursor, and others. You then create and use pipelines directly from your chat, and everything runs locally on your machine.
Do I need new API keys?
You bring your own LLM provider keys. Your pipeline runs go to your providers under your keys — DeeWaanAI orchestrates, it doesn't intercept.
Is my pipeline IP protected if I share it?
During evaluation, yes: the judging panel only ever receives the published test inputs and outputs, never your prompts, and nothing is used for training. But sharing itself is open source in the literal sense — anyone who forks your listing receives the pipeline definition itself. Publish only work you are willing to give away.
What does it cost?
Nothing. DeeWaanAI is free and open source, and marketplace listings are shared free of charge. Your only cost is your own LLM provider usage: runs go out under your keys, so your provider bills you directly.
Can I share a pipeline with someone who uses a different AI app?
Yes. DeeWaanAI is an MCP server, so you each connect your own client — Kimi, Claude, Cursor, or any MCP-compatible app. You lend the pipeline to that one account: by default they can run it and keep the answer without ever reading your prompts, and their run uses their own API keys. Publishing stays with whoever created the work, whatever privilege the share carries.
What is Research Mode?
Research Mode adds batch runs (up to 100 inputs in parallel) and pipeline chaining (feed one pipeline's output into another). It ships in the box — turn it on with a feature flag in Settings.
What can I do for free?
Everything. DeeWaanAI is the complete product at no cost: the custom pipeline designer with IF/WHILE loops and quality gates, Human-in-the-Loop approval gates, the AI Auto-Designer, up to 3 of your own LLM providers in one pipeline, and publishing to the marketplace. There is no trimmed tier to grow out of.
What are IF/WHILE loops?
IF/WHILE loops let you build conditional and repeating workflow logic — included free, not a paid extra. A quality gate can check an agent's output and automatically loop back until it passes — or route to a different step based on the result.
What is the AI Auto-Designer?
The AI Auto-Designer lets you describe your goal in plain language — for example, 'turn a product idea into a launch plan' — and the platform automatically generates the optimal pipeline with agents, prompts, quality gates, and approvals. It is included free, like everything else.