Decision engine

Every agent, one at a time

no market yet

Market

Load datadrop a file here or click to browse

Market from Divera AI handpicked on the website

Browse markets on Divera AI →
Pick a market there and tap Open in Decision Engine: you land back here with it loaded. Nothing is downloaded; the data stays on the engine.

This market

No market yet.Open one from Divera AI to see who is in it.

Training

The app trains a model of every agent: what each one cares about and why they choose the brand they do.
size of each training step · 0.1 is safe
most steps the app tries per agent
stop once explained this well
above this error an agent is left out

Model configuration

The active configuration is named in every page header. Retrain after switching: the weights on disk belong to the configuration that produced them.

Model checking…

Embedded · runs inside the app

Any .gguf dropped in appears here. Files in LM Studio's folder are reused, not copied.

LM Studio · its server on port 1234

Start the server in LM Studio (Developer → Start server) or run lms server start.

Custom · any OpenAI-compatible server

Ollama, vLLM, or a hosted endpoint — anything that speaks /v1/chat/completions. Alibaba Model Studio: base https://dashscope-intl.aliyuncs.com/compatible-mode/v1, model e.g. qwen-plus, plus your key. The key stays in llm.json on this machine.