# Accounts, models, and decisions

Connect providers, choose models, and set up decision models for routing and assistance.

## Accounts

The Accounts tab groups connections by provider. Connect a supported subscription or API key. Name accounts so you can tell work and personal usage apart. Expand Usage and settings for account controls. A chat subscription does not automatically include API credit.

## Models

The Models tab lists each provider’s newest model of every family first, with the models you turned on. Older models fold under Other models and stay out of the composer picker until you turn them on. Local endpoints such as Ollama and LM Studio need a running server. Check the endpoint and model name if no models appear.

## Decision models

A decision model is a small, fast model that answers typed questions about your message. Connect one in Accounts, such as Clef on Cloudflare, GPT-6 Luna with an OpenAI API key, a System One endpoint, or a llama.cpp or Ollama server. A new model, URL, or key passes a connection test before it saves. Chats can also call connected decision models from codemode scripts.

## Routing and assistance

The Decisions tab has two parts. Model routing sends your message to the decision model you choose, which picks the answer model and balances across accounts. Use Test routing to inspect a choice. Assistance picks MCP servers and skills for each message, with its own decision model or the routing one. While assistance is on, the composer names the model, such as Clef Flash assists, and you can turn it off for one message. A timeout or a low-confidence answer adds nothing.

## Capacity

Open Capacity beside the model picker to see each subscription account and its remaining allowance. Reset times appear when the provider supplies them. An unavailable allowance is unknown, not zero. API charges follow your provider account.