Agentary: Autonomous AI Agents in Your Browser

Agentary is an open-source framework that enables the execution of autonomous AI agents directly within the browser, offering a privacy-first solution for on-device AI processing. Designed for developers and privacy-conscious users, it runs entirely locally using quantized small language models powered by WebGPU or WebAssembly, eliminating server dependency, latency, and data leakage. Agentary supports prompt definition, reasoning, tool use, and local execution, making it ideal for applications like creative writing, interactive learning, local document Q&A, email assistance, and embedded customer support. With features such as browser-native execution, agentic workflows, streaming generation, multi-model support, and tool & API integration, it delivers powerful AI capabilities offline and on-device.

Key Features:

• On-Device AI Execution: Runs AI agents directly in the browser without servers.
• Privacy-First Design: No data leaves the user’s device.
• WebGPU & WebAssembly Support: Optimized performance with GPU acceleration.
• Autonomous Agents: Supports reasoning, tool use, and local agent workflows.
• Offline Functionality: Works once models are downloaded.
• Multi-Model Support: Compatible with various language models.
• Browser-Native Execution: Built for modern browsers with WebGPU support.
• Streaming Generation: Real-time output for interactive experiences.
• Tool & API Integration: Enables integration with external services.
• Open Source & Free: MIT-licensed, free to use and modify.

Pricing: Agentary is free to use under the MIT License, with no cost for individual or commercial use.

Conclusion: Agentary is a powerful, privacy-preserving framework that brings autonomous AI agents to the browser, enabling developers to build secure, fast, and offline-capable AI applications with minimal setup and no server infrastructure.