01
Products
Independent commercial assets with their own market, pricing and identity.
BlackRiver AI / Applied intelligence
Intelligence, made operational.
BlackRiver AI engineers local models, infrastructure and software that move intelligence out of the abstract and into real products — private, performant and built to ship.
ExecutionLocal-first systems
ResearchOpen-weight model engineering
InfrastructureRouting, optimization and control
StandardProduct-grade delivery
BlackRiver operating surface
Signal / Model / Product
Local models
Routing layer
Automation
Product systems
Thesis
The brand stays broad. The products stay focused.
BlackRiver AI is built to own more than one product. Some work becomes software. Some work becomes infrastructure. Some work stays private until it is useful enough to deserve a name.
The company is the house. The products are the surfaces.
02
Infrastructure
Core systems that make AI usage easier to route, measure, govern and optimize.
03
Labs
Private exploration around agents, voice, local models and machine workflows.
Live product · On-device AI
BlackRiver Workspace
A private ChatGPT-like workspace for local conversations, documents, OCR, vision and audio with Gemma 4 E2B and E4B.
Live product
TokenOptimizer
AI cost intelligence and optimization for companies using LLMs at scale.
Infrastructure product
BlackRiver Gateway
AI routing infrastructure: one endpoint for providers, coding agents and enterprise AI tools.
Flagship local model · Public GGUF
QwiVer3.6-35B-A3B
BlackRiver's post-trained Qwen3.6 sparse MoE for coding, agents and long-horizon reasoning. Our daily driver: 35B total, ~3B active, 262K native context, vision and MTP preserved.
35B / ~3B active
262K context
Vision + MTP
Q2 / Q3 / Q4 / Q8
llama.cpp
Open-weight model research
DeepRiver V4 Flash
Two llama.cpp-ready derivatives of DeepSeek V4 Flash 0731, engineered through quantization-aware refusal-direction projection with audited Q8_0 tensor replacement.
Flash / Pro
284.3B class
GGUF
llama.cpp
Voice intelligence lab
OmniVoice
Multilingual text-to-speech and zero-shot voice cloning running locally through WebGPU. Parakeet automatically transcribes reference audio before generation, with no account or usage quota.
TTS
Voice cloning
Parakeet ASR
WebGPU
On-device lab
Bonsai
A private local language-model workspace accelerated by WebGPU. Chat directly in the browser without sending prompts to a server.
Local LLM
WebGPU
Private
Multimodal lab
Gemma 4 Multimodal
Private multimodal intelligence for text, images and audio, running entirely on your device with WebGPU.
Vision
Audio
Text
High-speed language lab
LFM2.5 230M
A compact Liquid AI language model built for extremely fast private inference and structured tasks in the browser.
230MTextQ4_0
Local image lab
Bonsai Image 4B
Private image generation with compressed binary and ternary diffusion models running locally through WebGPU.
Image4BLocal diffusion
Operating layer
Quiet infrastructure.
The most valuable AI systems usually do not look loud. They route traffic, expose hidden costs, reduce waste and make operations more predictable.
BlackRiver Gateway powers the routing layer. TokenOptimizer turns usage into cost intelligence. BlackRiver Labs spans private browser inference, native local runtimes and open-weight model engineering through releases such as QwiVer3.6-35B-A3B and DeepRiver V4 Flash.
Routing
One endpoint, many providers.
Cost intelligence
Measure what matters.
Agent operations
Control machine workflows.
Local AI
Own more of the stack.
Company
BlackRiver AI Ltd.
A UK company created to build, own and operate AI products and infrastructure.
Legal nameBlackRiver AI Ltd
Company number17291854
JurisdictionEngland & Wales
BaseLondon, United Kingdom
Public assetsWorkspace · TokenOptimizer · Gateway · BlackRiver Labs · QwiVer · DeepRiver
Contact
Build with BlackRiver AI.
Products, automation, infrastructure, audits, partnerships and private projects.