How to choose models and routing rules

This page is for admins who want to decide which model does which work. Steps run on Claude Code or Codex, on Anthropic, OpenAI or local models (Ollama, LM Studio). When you are done, routing rules pick the model per step, and fallbacks keep runs going when a model hits a rate or usage limit.

Before you start#

• You are signed in as an admin.
• Claude Code (claude) is installed and logged in.
• Optional: the Codex CLI, for Codex steps. The shipped flows use it for reviews.
• Optional: Ollama or LM Studio, for local models.

Steps#

1. Check the coding agents#

Open Administration → Models. The Coding agents panel shows whether Claude Code and Codex are installed and logged in. To add Codex:

npm i -g @openai/codex
codex login

Codex uses your ChatGPT login. Claude Code uses your Claude login or an API key.

2. Learn the model specs#

Wherever you pick a model (a step, a flow's defaults, a routing rule, an eval), you write a model spec:

3. Add a provider and try a model#

Anthropic, OpenAI, Ollama (localhost:11434) and LM Studio (localhost:1234) are built in. To use another Ollama host or any Anthropic-compatible API, use Add a provider. Then use Try a model to check that a model works before you rely on it.

Local models cost nothing and keep your code on your machine. Choose models trained for coding and tool use, such as qwen3-coder or gpt-oss. General chat models often fail to call tools.

4. Set routing rules#

Rules decide which model runs which step. The first matching rule wins, even over models written in flows and blocks. A rule matches on a step id pattern, a flow name pattern and/or a visit number (From visit 2 means only retries). The presets add common rules:

• Retries on Opus: fix steps use Opus from their second attempt.
• Codex reviews Claude's code: review steps run on Codex.
• Small jobs on a local model: triage and learning steps run locally.

5. Set the default model#

Without a matching rule, a step uses its own model, then the flow's default, then the Default model. In a flow, set the default for all agent steps:

defaults:
  model: claude-sonnet-5-5
  agent: claude

A single step can override it with model:, agent: or provider:.

6. Add fallback models#

Fallback models are tried in order when a model hits a rate or usage limit. With continue on the first free one when a budget runs out, runs switch to the first free fallback (local or ChatGPT) instead of pausing when the daily or run budget is used up.

Check that it worked#

• The Coding agents panel shows each agent as installed and logged in.
• Try a model gives a result for the models you use.
• On a run page, the label next to each step name shows which agent and model ran it. Check that it matches your rules.

Next steps#

• Control costs and read the dashboard
• Write your own flow
• Guide: Models, agents and routing
• Guide: Providers and local models