The Agentic Web
Begins at the Edge

Agentary is an open-source framework that lets developers run autonomous AI agents directly in the browser

AI doesn't need to live in the cloud

Agentary brings AI execution directly to the browser, letting every website host its own autonomous agents

No servers, no latency, no data leaving the user’s device. Just fast, private, local intelligence

How Agentary Runs Agents in Your Browser

Agentary runs quantized small language models entirely on-device, orchestrating reasoning and tool use without external servers

💭

Prompt

Define a goal for your agent, from simple prompts to multi-step workflows that coordinate tools and reasoning.

🧠

Run

A quantized model loads directly into your browser and runs on WebGPU. No servers, no API calls, no data leaving the client.

🤖

Think

Agents reason, plan, and manage context locally, deciding what to do next without human intervention.

Act

Let your agent call tools, queries APIs, and returns results, all executed locally on your device.

Frequently Asked Questions

Everything you need to know about building with Agentary

Agentary is a local-first SDK that enables in-browser AI agents using WebGPU. It allows developers to build autonomous AI experiences that run entirely on the client-side, with no data sent to servers. This provides privacy-first, serverless AI capabilities directly in web applications.
Agentary runs inference fully on-device with no data sent to servers. Unlike cloud-based AI solutions that require API calls and expose your data to third parties, Agentary processes everything locally in the browser using WebGPU or WebAssembly. This means faster responses, complete privacy, offline functionality, and no usage limits or API costs.
Yes! Once the models are downloaded, Agentary works completely offline. All AI inference happens locally in the browser using WebGPU or WebAssembly, so you can build applications that function without an internet connection.
Agentary works best in browsers with WebGPU support, including Chrome 113+, Edge 113+, and Opera 99+. For browsers without WebGPU, Agentary automatically falls back to WebAssembly (WASM) for broader compatibility, though with reduced performance.
Yes, Agentary is completely free and open-source, released under the MIT License. You can use it in personal and commercial projects without any licensing fees, API costs, or usage limits.
Install Agentary via npm with "npm install agentary-js", then import the SDK in your JavaScript or TypeScript project. Check out our documentation at docs.agentary.ai and explore examples on GitHub to see how to create your first browser-based AI agent.

Ready to Get Started?

Join developers building the future of browser-based AI