Open-source engine for faster local language model inference with bit-for-bit identical output
Stage
MVP
Not given
Founded
Not given
Not given
Based in
FR
Not given
Pricing model
Paid
Open source
Pricing
Pay-as-you-go credits. Top-up fees: Business rate 5.5% + €0.30, Standard rate 6.5% + €0.30, International 8.5% + €0.30. Minimum top-up €5.00. Tokens billed at upstream provider's per-token price in EUR with no per-token markup.
Not given
Key features
OpenAI-compatible API, works with existing SDKs and tools without modification
Per-request carbon emissions tracking and grid carbon intensity reporting
Multi-provider routing with choice of provider, model, and region per request
Model aliases for repointing requests without code changes
Auto-routing to select lowest-carbon route based on policy
Jurisdiction filtering including EU-only provider option
Usage dashboard with per-model, per-provider, and per-key breakdown
CSV export for auditable per-request usage and carbon data
One invoice for all providers with automatic EUR conversion
No lock-in with base URL configuration allowing instant provider switching
Speculative decoding for faster token generation
Prompt caching for context reuse
Real-time prompt processing while typing
Concurrent request batching
Bit-for-bit identical output verification
Multi-GPU and multi-backend support (CUDA, Vulkan)
Vision model support with image encoding
Mixture-of-experts model support
RAD-1 automatic draft token generation
Cross-platform support for Windows and Linux
What makes it different
Sovereign EU hosting outside CLOUD Act reach with GDPR compliance and your choice of provider jurisdiction per request
No data training or retention; prompts forwarded unchanged to chosen provider only
Exact deterministic inference: all default optimizations preserve bit-for-bit identical output compared to plain decoding
4.5× faster performance on editing tasks on small models compared to llama.cpp
Verifiable reproducibility through receipt system that records all request parameters
Nine optimization techniques enabled by default with ability to disable each individually
No dependency on draft models or external selections for speculative decoding
Kinetic