Models and private inference
The reasoning layer for agents, with explicit controls over model processing.
Frontier and open-source models
Seer provides access to a suite of frontier and open-source LLMs through its inference layer. Agents can use a model suited to their task while drawing on the same financial tools and account controls. The model catalog identifies supported capabilities, context limits, pricing and available processing routes.
Model choice depends on the work: reasoning over a long filing, coordinating several research tools and classifying incoming events have different requirements. Tool support, response quality, latency and cost all matter. Open-source describes the model’s availability; it does not, by itself, establish where inference runs or how data is retained.
What private inference covers
Private inference controls how model inputs and outputs are processed. Each route identifies the processing boundary and its retention and training-use policy. Seer applies the selected policy before dispatching model context. Tool credentials remain outside the prompt.
Model processing is separate from saved workspace history. Conversations, research artifacts and financial records follow their own storage and access settings. Private inference does not make a public blockchain transaction private or change the data required by a connected financial service.
The model processing boundary
Language models, tools and Aura
The language model interprets the task, selects tools and explains results. Financial tools retrieve evidence or run defined calculations. Aura produces quantitative forecasts for specified outcomes and horizons. An agent combines these capabilities without treating a generated opinion, an observed price and a numerical forecast as interchangeable.
The sources and calculations behind a result remain inspectable. A model’s ability to explain an investment does not grant authority to execute it.
External agents
An agent connected only to Seer’s tools can keep its existing model provider. It receives tool results through its integration and applies its own processing and storage policies to those results. When an agent uses Seer’s model services, the selected Seer inference route governs those model calls.