Packages

Shared candle primitives (triplane sampling, marching cubes, gaussian splat types, DINOv2 wrapper) reused across Blazen 3D backends (TripoSR, TRELLIS, Hunyuan3D)

Capability-agnostic shared types and traits for Blazen's audio engine crates (TTS, STT, music, codec, voice conversion)

Neural audio-codec backends for Blazen — encode PCM to discrete codebook tokens and back for generative audio models

Shared candle primitives (DiT attention, SwiGLU, AdaLN-Zero, RoPE, Fourier features) reused across Blazen audio + 3D diffusion backends

Vendored fork of candle-transformers' Descript Audio Codec (DAC) model, patched for Blazen

Music + sound-effect generation backends for Blazen (MusicGen, AudioGen, Stable Audio)

Local text-to-speech backend for Blazen using Piper (ONNX Runtime)

Vendored fork of piper-rs (Piper neural text-to-speech) patched for the Blazen workspace

Multi-backend speech-to-text engine crate for Blazen (whisper.cpp, candle)

Multi-backend text-to-speech crate for Blazen (any-tts local engines + OpenAI-compatible HTTP)

Multi-backend voice-conversion crate for Blazen (RVC and related speaker-transform engines)

Local speech-to-text backend for Blazen using whisper.cpp

Hand-rolled C ABI over blazen-uniffi for the Ruby gem (via cbindgen + FFI gem) and any other FFI host

CLI for the Blazen workflow engine

Distributed workflow control plane for Blazen — workers connect inbound to a central server

Core workflow engine for Blazen - event-driven, async, pausable workflows

Embedding backend facade for Blazen: picks fastembed (ORT) or tract (pure Rust) per target

Local embedding backend for Blazen using HuggingFace candle

Local embedding model backend for Blazen using fastembed-rs (ONNX Runtime)

Pure-Rust ONNX inference backend for Blazen embeddings (tract-onnx)

Event traits and built-in event types for the Blazen workflow engine

Vendored ggml C library, built once and reused by Blazen's whisper-rs-sys / llama-cpp-sys-2 / diffusion-rs-sys consumers via patch.crates-io.

Local image generation backend for Blazen using diffusion-rs (pure Rust Stable Diffusion)

LLM provider abstraction layer for the Blazen workflow engine

Local LLM backend for Blazen using candle inference engine

Core LLM primitives for the Blazen workflow engine (traits, HTTP, providers, content)

Local LLM backend for Blazen using llama.cpp inference engine

llama.cpp HTTP-server (llama-server) proxy provider for Blazen — OpenAI-compat + native completion + runtime LoRA adapter toggle

LM Studio OpenAI-compat + native model load/unload proxy provider for Blazen

Local LLM backend for Blazen using mistral.rs inference engine

Ollama HTTP proxy provider for Blazen — generate/chat/embed + Modelfile-based LoRA adapter mounting

HuggingFace Text Generation Inference (TGI) HTTP proxy provider for Blazen — native /generate + OpenAI-compatible /v1/* with per-request adapter selection

vLLM OpenAI-compat + LoRA adapter management proxy provider for Blazen

Shared option + request types for Blazen local LLM backends (candle, mistralrs, llama.cpp) and the controlplane local-model factory seam

Derive and attribute macros for the Blazen workflow engine

VRAM-aware model manager with LRU eviction for the Blazen workflow engine

Shared leaf crate of media request/result/error DTOs for Blazen's 3D, music, and voice-conversion capability surfaces

Memory and vector store for Blazen with ELID integration

Valkey/Redis backend for blazen-memory

Shared model download and cache layer for Blazen local-inference backends

Small runtime-environment probes for Blazen — outbound IPv4/IPv6 egress-preference detection (curl/AI_ADDRCONFIG-equivalent) for the reqwest/tonic clients, and CUDA-driver availability detection to guard candle/cudarc device probes on hosts without CUDA.

Distributed workflow execution peer for Blazen — gRPC server + client over tonic

Checkpoint and persistence storage for Blazen workflows

Multi-workflow pipeline orchestration for Blazen

Prompt template management for the Blazen workflow engine

Generated blazen-llm provider wrapper for AtlasCloud (atlascloud).

Deterministic ProviderSpec -> blazen-llm provider wrapper code generator

Generated blazen-llm provider wrapper for Cohere (cohere).

Generated blazen-llm provider wrapper for DeepSeek (deepseek).

Generated blazen-llm provider wrapper for Fireworks AI (fireworks).