Model card
Rivier One.
Makes every frontier AI feel like one. Carries your memory across them. Trained by Rivier.
Summary
Rivier One makes every frontier AI model feel like one. It carries your memory and intelligence across the assistants you use, picks the right model for each request, and answers itself when it can. Trained by Rivier.
It powers every Rivier surface — the assistant inside the AI apps you already use, the agents that act on your behalf, and the Rivier app on rivier.ai.
Intended use
Orchestration inside Rivier. Rivier One reads each request together with grounded memory, picks the right intelligence, and either answers directly or routes to a stronger model when the work earns it.
Rivier One is embedded in the Rivier product. It is not a general-purpose chat replacement.
Capabilities
Structured task routing across the open and named-model tiers.
Grounded answer quality on user memory recall.
Instruction following and tool-use intent.
Per-request cost, latency and privacy ceilings as routing inputs.
Limitations
Text-only. Multimodal requests route to a stronger model rather than being handled directly.
No multi-step planning or task decomposition. Multi-round orchestration is the role of Rivier Two.
Training
Rivier One is trained by Rivier using a Rivier-specific orchestration objective. Training data includes publicly available preference data and Rivier-constructed prompts mirroring real platform usage.
Weights, training data, training pipeline and reward function are proprietary. Architectural details are not disclosed.
License
Rivier One weights are proprietary. Available exclusively to Rivier users via Rivier's product surfaces.
Enterprise deployments under restricted commercial license are evaluated case-by-case.
Versions
Rivier One — small orchestration model. Currently shipping.
Rivier Two — multi-step orchestration model. In development.
Contact
Research and benchmark inquiries: research@rivier.ai
Enterprise deployment: enterprise@rivier.ai