Lemonade by AMD: A Unified API for Local AI Developers

The AI PC Ecosystem for Developers, by Developers

Developers need free, private, and optimized on-device AI with:

• All the LLM, speech, and image capabilities required for natural interactions and powerful outcomes.
• Zero-friction deployment across the entire AI PC ecosystem, especially AMD Ryzen™ and Radeon™ products.

A growing community of open-source developers and AMDers is building Lemonade to make this a reality!

The Real Developer Challenge: Universal AI PC Deployment

For any developer working with local AI, the challenges are familiar: a mix of hardware architectures, competing runtimes, platform-specific optimizations, and deployment hurdles. The result is a fragmented ecosystem. These friction points slow down not just individual developers, but the broader AI PC ecosystem.

That's why Lemonade prioritizes the ecosystem’s needs:

• Implemented in lightweight C++ for native and quick installation across Windows, Linux, MacOS and Docker.
• Automatically configures optimized NPU, GPU, and CPU backends for text, image, and audio inference for each specific computer.
• Ability to run multiple models simultaneously, limited only by available RAM.

Multi-Modal AI: Beyond Just Large Language Models

Modern AI applications extend well beyond text, and Lemonade is designed to support multiple generative modalities through a consistent, local‑first developer experience.

• Text generation for natural language interfaces, agents, and tool‑driven workflows.
• Image generation for content creation and visual feedback loops inside applications.
• Speech‑to‑text and speech recognition for accessibility scenarios and hands‑free interaction.

Multi-Engine Architecture: Performance That Scales

Lemonade’s multi‑engine design lets developers focus on application behavior rather than hardware details. With Lemonade, developers interact with a unified interface that supports text generation, image synthesis, speech recognition, and text‑to‑speech across different hardware accelerators. Lemonade handles backend selection and optimization based on available hardware, reducing integration complexity while preserving flexibility.

Engine support is ever-expanding and already includes: llama.cpp, ryzenai, FastFlowLM, whisper.cpp, stablediffusion.cpp, and Kokoro.

Lemonade Desktop App

Beyond the command‑line interface, Lemonade includes a full‑featured desktop application that serves as a central control center for local AI development. The app provides a graphical interface for downloading and managing your model library, with a visual Model Manager for browsing, installing, and organizing models. In addition to the built‑in chat experience, the UI exposes interfaces for working with common AI workflows such as image generation, speech input and output, embeddings, and reranking. This makes it easy to test models, switch configurations, and monitor server status from a single place during development. 

The desktop app also handles the complexity of backend selection and hardware optimization automatically, making it accessible to both developers and non-technical users. Whether you're experimenting with new models, building AI workflows, or want a user-friendly way to run LLMs locally, the desktop app lowers the barrier of entry from a developer tool into a comprehensive AI platform for your PC.

Community-Driven, AMD-Optimized

Lemonade balances two critical needs: community alignment and enterprise reliability.

The project is community driven, shaped by developers who see value in the approach and choose to contribute to it. Our roadmap is based on actual user feedback, integration requests, and ecosystem needs. For example, when a developer said, "I need text-to-speech," Michele Balistreri (@bitgamma) built support for text-to-speech. We seek out our developer community to guide us to impactful actions and Lemonade is shaped by developers who see value in the approach and choose to contribute to it.

With AMD backing Lemonade, the stack is thoroughly tested and optimized for AMD hardware. Providing developers with confidence that their applications will work consistently across AMD's expanding AI PC hardware lineup, from Ryzen AI NPUs to integrated and discrete GPUs.

Developer Ecosystem Integration

To validate our goal of making Lemonade easy to integrate, the team put it to the test by bringing it into the tools developers already rely on. Lemonade now works with:

• VS Code via GitHub Copilot integration.
• Continue for code completion.
• n8n for workflow automation.
• OpenWebUI for a feature-rich, self-hosted AI platform.
• And more…

These integrations exist today because Lemonade's OpenAI compatibility makes integration straightforward for tool maintainers.

Come Build with Us

With Lemonade, running AI locally is straightforward thanks to OpenAI API compatibility. If your app already targets OpenAI, in the cloud or via another local runtime, moving to Lemonade is a simple change: update the endpoint and select a model. From there, the same code runs locally with AMD‑optimized CPU, GPU, and NPU backends, giving you privacy by default and predictable costs. If this is the direction you want the ecosystem to go, Lemonade is where that work is happening. Come build with us!

To Get Started

• Check out the Lemonade Website to learn more and download.
• Check out GitHub repo here.
• Join Discord here.
• Join the AMD AI Developer Program.

Subscribe to the Ryzen AI Newsletter to get the latest tools and resources for AI PCs.