Jump to a section

Free · open source · bring your own key

Your AI workspace lives in your browser.

Bring your own key. Chats, files, and knowledge stay in your browser; the server stores no provider keys.

Sign in with ChatGPT or paste an API key. No account to create.

ai.diy workspace

Preview with sample data. You ask the assistant to chart a CSV. It runs Python in your browser tab, answers, and saves a chart and a cleaned file to Canvas. The chat is saved in your browser and the server stores no provider keys.

Illustrative playback with sample data. The real demo is one click away.

Your keys stay under your control.

Inspect where a request is read, forwarded, and saved. Nothing here is a persistent server-side key store.

LOCAL DATA PLANEINSPECT

OWNED

Your browser

  • Provider keys stay in the browser and are relayed per request.
  • When Web Crypto and IndexedDB are available, settings are encrypted at rest with AES-GCM. The payload is in localStorage; the envelope key is in IndexedDB. Otherwise storage may be plaintext.
  • Threads, Canvas artifacts, memory, knowledge, and usage events persist in IndexedDB.

TRANSIT

Node relay

  • No persistent provider keys. The server does not need LLM keys in its environment.
  • Each request is forwarded to the provider you select.
  • Private-network provider and MCP URLs are rejected unless a trusted self-host opts in.
  • Remote MCP connections reject redirects.
  • Optional sliding-window rate limits via RATE_LIMIT_RPM.

CHOSEN

Chosen provider

  • Cloud or local endpoint you choose — 26 integrations, including Ollama and LM Studio.
  • Retrieved knowledge-base context may be sent to the selected cloud model.
  1. Key read from browser
  2. Relayed per request
  3. Provider response streamed
  4. Saved to IndexedDB

Bring any model into the same thread.

Connect 26 cloud and local providers, then switch models without moving your workspace or copying context between apps.

Also supported: Anthropic, Groq, Mistral, Bedrock, Azure, Vertex, Together, Hugging Face, LM Studio, and custom OpenAI-compatible endpoints. No key to paste? Sign in with a ChatGPT, Grok, or Kimi plan instead (beta); those sessions live in an encrypted server session, not in browser storage.

Everything in one workspace.

Create files, run code, search the web, and connect your apps with the models you already use.

Canvas

Artifacts open beside the thread: text, HTML previews, images, and Python output. They are saved with the chat in IndexedDB.

Python in the tab

Pyodide runs in your browser, not on a server. Charts and files land in Canvas.

Knowledge base

On-device WASM embeddings and search. Documents are not uploaded to a vendor vector store; retrieved context may be sent to the model you choose.

Tools and MCP

Keyless web search (Firecrawl, Parallel), URL fetch, a calculator, remote MCP servers, and skills. You choose what the model may use.

Connected apps

Authorize Gmail, GitHub, Notion, Slack, and more through Composio with your own key. The assistant acts only in the tools you connect.

Backups you control

Export your chats, or back up to S3, WebDAV, or Google Drive straight from the browser. Off until you turn it on.

Self-host in one command.

A Node production build, Docker Compose, or a Vercel preview.

Deploy to Vercel

Frequently asked questions.

The ownership model, supported providers, browser storage, and the self-hosting path.

What is ai.diy?

ai.diy is an open-source, self-hosted AI workspace for chat, research, tools, Canvas artifacts, memory, and local knowledge. It uses bring-your-own-key (BYOK) access so you choose the provider and model instead of being locked into one hosted AI service.

Are my AI provider keys stored on the ai.diy server?

No. Provider keys stay in the browser and are relayed per request. The server does not keep them as persistent secrets. Self-host when you need control of the network boundary.

Where does ai.diy store chats and documents?

In the browser. Threads, files, Canvas artifacts, memory, knowledge chunks, usage events, and preview sessions use IndexedDB. Settings use localStorage, encrypted with AES-GCM when Web Crypto and IndexedDB are available. Optional S3, WebDAV, and Google Drive backups are client-side and off until you enable them.

Which AI providers work with ai.diy?

ai.diy supports 26 integrations: OpenAI, ChatGPT subscription, Grok subscription, Kimi subscription, GLM, MiniMax, Anthropic, Gemini, Groq, Cerebras, Fireworks, Perplexity, Cohere, OpenRouter, xAI, DeepSeek, Bedrock, Azure, Vertex, Vercel Gateway, Together, Mistral, Hugging Face, Ollama, LM Studio, and custom OpenAI-compatible endpoints.

Can I self-host ai.diy?

Yes. Run the production build on a standard Node.js server or use Docker Compose. The server acts as a request relay and does not need provider API keys in environment variables. You can open the hosted demo first or deploy the MIT-licensed source code on infrastructure you control.

Does ai.diy replace my AI provider or pay for model usage?

No. ai.diy is the workspace layer. You bring authorized provider keys or connect local models such as Ollama and LM Studio. Provider pricing, quotas, availability, and data policies remain controlled by each provider, and any usage charges are yours.

Your models. Your data. Your machine.

Open the workspace, drop in a key, and start thinking. Or clone the repo and self-host in one command.