TensorSharp

TensorSharp runs local AI models for .NET developers and people who want chat, media generation and agent tools on their own hardware. You can embed it in a C# application or use its CLI, browser interface or companion app, TensorAgent.

• Text and multimodal chat: Work with models from families including Gemma, Qwen, DeepSeek and GLM, with image, audio and document input where the model supports it.
• Image and video creation: Generate and edit images with Qwen-Image 2.1, including masked edits and compatible LoRA plug-ins, or create video with LTX 2.5, MiniMax-H3 and Wan; LTX and MiniMax-H3 can also generate stereo audio.
• Personal knowledge collections: Search imported files, folders and saved chats, retrieve cited passages and open retained originals, with optional embeddings for semantic search.
• Agent workflows: Give agents file, shell and document tools, use Agent Skills and delegate bounded tasks to sub-agents.
• Application integration: Serve chat, Responses and embeddings through Ollama- and OpenAI-compatible APIs, using embedding models such as EmbeddingGemma 2, Arctic Embed and MiniLM.

TensorSharp is free and open source under the BSD-3-Clause license. The engine runs on Windows, macOS and Linux, using CPU inference or supported CUDA, Vulkan, Metal and MLX backends. TensorAgent has desktop builds for Apple Silicon Macs and Windows x64; iPhone and iPad require source builds. Inference runs locally without a metered API. Model weights download separately, and community skill discovery needs network access. Model, input and acceleration support vary by backend and model family.