Voyager SDK | Axelera AI

End-to-end integrated software stack

Purpose-built for AI inference, the Voyager SDK moves you from idea to deployment quickly, with deep customization when you need it. It supports applications built on proprietary or standard industry models, and its end-to-end integration and API compatibility with de-facto industry standards mean high-performance AI on Axelera hardware, without the slowdown.

Voyager SDK is available for download on our Docs Portal.

Model Zoo

Optimized deployment

The Voyager SDK makes it easy to build high-performance AI applications. You describe your end-to-end pipeline in YAML, a simple, human-readable, high-level declarative language, and the SDK takes care of the rest. With Voyager SDK, you can:

Define complete pipelines in YAML, including one or more neural networks, plus their pre/post-processing tasks, which can include sophisticated image processing operations
Skip the manual optimization work. The SDK automatically compiles, optimizes, and deploys your entire pipeline, making use of host processors (CPU, embedded GPU, or media accelerator) and AIPU as required
Build across a wide range of host architectures and platforms to meet the requirements of diverse application environments
Go from pipeline to solution fast. Flexibly embed a pipeline into an inference service to get a variety of preconfigured, out-of-the-box solutions, from single-stream use cases to distributed processing of multiple 4K streams

Model Zoo

Join the Community

Customer models	Model zoo		Non-neural libraries

Low-code deployment flow

Gstreamer	TVM compiler (PyTorch, TF, ONNX)		Open CL/GL compiler

Pipeline Callback API

Customer application			Inference server

Metis runtime
↑ PCIe ↓		↑ Ethernet ↓
Metis hardware

An AI coding agent that's fine-tuned for Axelera hardware

Voyager Wingman has deep knowledge of our SDK and documentation, for developers who want to build production-grade solutions faster. It builds, debugs, and optimizes AI inferencing pipelines using natural language inputs for 20 to 30% higher accuracy than general-purpose AI coding tools.

Every developer needs a wingman.

Try Wingman Now

Expert Assistance

Build end-to-end applications. Describe the pipeline you want, from a single model to a multi-stage cascade, and get a working result while preserving accuracy.
Optimize performance. Get concrete recommendations on compiler configurations and how to get the most out of Axelera hardware.
Debug faster. Diagnose configuration errors, device enumeration issues, and models that fail to compile, with guidance toward a working result.
Get documentation on demand. Ask about supported operators, APIs, and runtime behavior, and get contextual answers with examples instead of searching reference docs.

Available as a web-based chat and a standalone app, with a plugin for popular AI coding frameworks coming soon. Its knowledge updates automatically with every Voyager SDK release, so there's nothing to download or maintain.

Try Voyager Wingman free through the Axelera Developer Community with a starter credit allowance, or use the app at no cost with your own API key.

Try Wingman Now

Introducing Voyager Wingman by Axelera AI

Welcome to the future of AI engineering. In this video, Steven Hunsche (Senior AI Solutions Engineer) introduces Voyager Wingman, an innovative application designed to automatically create end-to-end AI pipelines for Axelera hardware. By mapping and understanding Axelera's entire codebase, Voyager Wingman works alongside LLMs like Claude and ChatGPT to design, build, and test custom pipelines in minutes—no deep machine learning expertise required. This is true "vibe coding" brought to life.

How to Build a Soccer CV Analytics App in Minutes with Voyager Wingman!

Voyager Wingman builds a real-time soccer broadcast console with computer vision, just from a prompt. No pipeline code from scratch: Voyager Wingman generates the full workflow using the Voyager SDK and Axelera's documentation, getting max hardware performance in minutes, not hours.

Track player velocity, map tactical heatmaps, or analyze key moments from Dallas, Atlanta, or MetLife Stadium; Voyager Wingman makes it seamless.

Build a Real-Time Endoscopy AI Analyzer with Voyager Wingman

Voyager Wingman builds a real-time endoscopic video analysis pipeline for the edge, just from a prompt. No manual coding, no driver debugging, no model tuning: fully optimized, hardware-accelerated code in minutes, using the Voyager SDK and Axelera's hardware knowledge.

AI acceleration at the medical edge, without the software bottleneck.

Voyager Toolkit

Everything you need to build, faster