Software Mansion

Software Mansion AI Labs

We're an AI lab building small, fast and specialized models as well as agentic systems for business. We handle the full lifecycle, from training to deployment and observability, with a focus on local and hybrid inference, including edge devices.

This organization hosts our own model weights, along with models prepared for use in our projects, converted from PyTorch for runtimes and backends including ExecuTorch, XNNPACK, Core ML, MLX, Vulkan, QNN, and WebGPU. They cover a wide range of tasks, from vision and speech to multimodal LLMs and embeddings. Browse our collections below to find models by task, or try out the Spaces section to see our tech live.

Open-source developer tooling

• React Native ExecuTorch: Run 70+ AI models on-device in React Native apps, built on ExecuTorch.
• ruNNtime: An AI runtime built on top of TypeGPU.

Models and products

• Velonos: On-device redaction that keeps sensitive data out of AI tools.
• Mailboy: An AI agent that reads papers, repos and articles and sends you a personalized AI digest.
• Private Mind: A free, private AI chat app that runs entirely on your phone

Got an AI problem? Let's solve it together.