mrx: one session, three models, three different jobs
Assign a reasoning model to think, a cheap model to write, and a third to call tools — mixing Ollama, OpenAI, Anthropic and OpenRouter in the same conversation.
Everything filed under “cli”.
Assign a reasoning model to think, a cheap model to write, and a third to call tools — mixing Ollama, OpenAI, Anthropic and OpenRouter in the same conversation.
Most AI CLIs run one agent on one expensive model. Cascade runs a three-tier hierarchy that plans, delegates and executes in parallel — routing each step to the cheapest model good enough for it.