An AI coding assistant you actually own.
Amplify is a free, open-source, self-hostable AI coding assistant. Bring your own LLM keys, run it in your browser, and keep every chat, file, and snapshot on your own machine.
Closed assistants lease you a model and a bill. Amplify is the opposite - open source, self-hosted, twenty-two providers behind one chat. Bring your keys. Run it in your browser. Own every byte.
Why choose Amplify.
What changes when your AI coding assistant runs on your terms — your models, your runtime, your data.
22+ Providers
OpenAI, Anthropic, Gemini, DeepSeek, Ollama and more behind one chat. Swap models mid-conversation without losing context.
Self-Hostable
Docker, Cloudflare Pages, Coolify, or bare metal. MIT-licensed, no per-seat fees, no markup on your API spend.
Own Your Data
Every chat, file, and snapshot lives in your browser's IndexedDB. Nothing reaches a vendor cloud unless you connect it.
Real Sandbox
A browser-native WebContainer runs Node.js, installs packages, and executes your code with a live preview. No remote VM.
Token-Efficient
Sub-chat workers and reusable memory keep context small. New chats start with ~1,500 targeted tokens, not your whole history.
Open Source
Fork it, audit it, commercialize it. MIT-licensed core, with Skills and MCP extensibility built in.
From bolt.diy to Amplify.
Amplify began as a fork of bolt.diy — the community-maintained, open-source AI coding assistant. bolt.diy itself was only possible because StackBlitz open-sourced their WebContainer architecture: a full Node.js runtime running entirely inside the browser, with no remote VM required. That act of generosity is the technical foundation everything here is built on, and we owe them a genuine debt of credit.
We inherited bolt.diy’s multi-provider LLM orchestration, browser-native sandbox, and MIT-licensed core — then kept building. Sub-chat workers to keep context lean, a reusable vector memory layer, 8 built-in skills and 114 design systems,, first-classMCP server support, and 22+ model providersbehind one interface. Today Amplify is an independent, self-hostable coding assistant: open source at the core, zero per-seat fees, and every byte of your data staying on your machine. The DNA hasn’t changed — we just kept the promise.
FAQ.
Got questions? We've got answers. Here's everything you need to know about Amplify, our 22+ provider integrations, and how self-hosting actually works.
Amplify is a free, open-source, self-hostable AI coding assistant that runs entirely in your browser. It combines a full IDE-like environment (code editor, terminal, file explorer, live preview) with an AI assistant that can read files, write code, search the web, and execute commands — all powered by your own LLM API keys.
Amplify supports 22+ providers out of the box, including OpenAI, Anthropic, Google Gemini, Hugging Face, Mistral, DeepSeek, xAI, Groq, Cohere, Fireworks AI, Perplexity, OpenRouter, Together AI, Cerebras, Moonshot, Hyperbolic, Z.ai, GitHub Models, AWS Bedrock, Ollama, LM Studio, and any custom OpenAI-compatible endpoint. You can switch between them mid-conversation without losing context.
No. Amplify is MIT-licensed and self-hosted. Your API keys go directly from your browser to the provider's API — they never pass through Amplify's servers because there are no Amplify servers. You only pay the provider you already use, at their normal rates. There is no per-seat SaaS fee and no markup.
Each provider's own platform is locked to that one vendor's models, has no real code execution sandbox, has no MCP or Skills extensibility, and stores your conversation history on their cloud. Amplify unifies 22+ providers behind one chat surface, runs real Node.js in a browser-native WebContainer, supports 8 built-in skills and 114 design systems, and any MCP server, and keeps every chat, file, and snapshot in IndexedDB on your own machine.
Yes. Conversations are stored independently of providers. When you switch models, only the API endpoint changes — the new provider picks up the existing conversation as context. You can use Claude for hard reasoning, switch to Groq for fast boilerplate, and finish with a local Ollama model for sensitive code, all in the same chat.
Six deployment methods are documented: Docker (production, development, or prebuilt image from GitHub Container Registry), Cloudflare Pages (edge-hosted with COEP/COOP headers), Coolify, an Electron desktop app (macOS, Windows, Linux), native bare-metal on Node.js 18.18+, or behind a reverse proxy (nginx or Caddy) for SSL. Memory requirement is 4GB minimum, 8GB recommended.
Everything stays on your machine. Chat history, file snapshots, project commits, and screenshots live in an IndexedDB database called amplifyHistory. The BM25 vector store (used for cross-session memory) lives in a separate IndexedDB database called amplify_vector_stores. Nothing is uploaded to a vendor cloud unless you explicitly connect a cloud integration like Supabase, Vercel, or Netlify.
Yes, Amplify is MIT-licensed and fully open source — fork it, audit it, modify it, commercialize it. One caveat: the WebContainer sandbox runtime (used for in-browser code execution) is proprietary to StackBlitz and requires a commercial license for production use. If you migrate the sandbox layer to your own Docker container or any other container runtime, the entire stack is yours to commercialize under the MIT license.
Tools like Bolt, v0, and Lovable run everything in one long chat, so the context window grows linearly and the AI gets slower and noisier over time. Amplify uses sub-chat workers (each task gets its own isolated context), reusable memory (key info is extracted to vector stores after each chat), and summarization at the 70% threshold. New chats start fresh with only ~1500 tokens of targeted context injected — not the full baggage of every previous message.
No. Amplify does not rate-limit you, throttle your requests, or set ceilings below what your provider allows. It reads the per-provider RPM/TPM/RPD limits you've configured in Settings and adapts via a Pre-Flight Check, RPM Throttling, and Auto-Shrink-to-TPM to avoid triggering 429 errors from your provider.
Yes. Ollama (default http://127.0.0.1:11434) and LM Studio (default http://127.0.0.1:1234) are first-class providers with unlimited rate limits. Use 127.0.0.1 instead of localhost to avoid IPv6 resolution issues. If you run Amplify in Docker, set RUNNING_IN_DOCKER=true so localhost is rewritten to host.docker.internal automatically.