Browser-resident Gemma 4 inference: pure Rust → WebAssembly + WebGPU. Loads Ollama's on-disk GGUF blobs and runs the forward pass on the local GPU via hand-written WGSL.
Local LoRA fine-tuning for the rullama Rust runtime, built on rullama's wgpu kernels. Compiles for native and wasm32; browser harness lives in examples/web.
Cloud fine-tune APIs (OpenAI, Anthropic, Together, Fireworks, Anyscale, Bedrock, Vertex AI) + dataset pipelines for the Brainwires Agent Framework. Local LoRA/QLoRA/DoRA fine-tuning and training-from-scratch live in the sibling `rullama` workspace (`rullama-finetune`, `rullama-training`).
DEPRECATED: moved to rullama-training in the sibling rullama workspace — see README
The Brainwires Agent Framework — build any AI application in Rust
Adaptive prompting techniques, K-means task clustering, and temperature optimization for the Brainwires Agent Framework
Speech (TTS / STT) provider clients for the Brainwires Agent Framework — Azure Speech, Cartesia, Deepgram, ElevenLabs, Fish Audio, Google TTS, Murf, plus browser-native `web_speech` (wasm32)
Codebase indexing + hybrid retrieval (vector + BM25) for the Brainwires Agent Framework. Includes AST-aware chunking via tree-sitter (12 languages), Git history search, and reranking via spectral diversity / cross-encoder.
Built-in concrete tools for the Brainwires Agent Framework — bash, file_ops, git, web, search, code_exec, semantic_search, browser, email, calendar, system. Built on `brainwires-tool-runtime`.
Tool execution runtime for the Brainwires Agent Framework — executor trait, registry, error taxonomy, sanitization, validation, transactions, smart router, plus optional orchestrator / OAuth / OpenAPI / sandbox / sessions / RAG-tool-search modules