Coding agents

Router is compatible with all coding agents that support the OpenAI Responses API. In addition, we provide direct integrations with several popular coding agents through the ramp router command in the Ramp CLI:

You can also set up any of these agents, and Cursor, by hand without running an installer. The manual setup guides walk through each one.

For Claude Code, Codex, or OpenCode, install the Ramp CLI, then configure Router models and connectors↗:

curl -fsSL https://agents.ramp.com/install.sh | sh -s -- --no-skills
ramp router configure --connectors

If the installer updates your PATH, open a new terminal before running ramp router configure --connectors.

Select the agents you wish to configure using the up and down arrow keys to navigate, Space to select or deselect an agent, and Enter to confirm. Then sign in with Ramp in your browser and approve connector access for the account and company shown. Setup offers to reuse a saved Router API key or creates one for you if you don't have one.

Connect your apps on Connectors↗, then restart your agent. If the page isn't available, ask your Ramp administrator about Router access. See Connectors↗ for company permissions.

You may also configure a specific agent by passing in its name, for instance ramp router configure opencode --connectors. Connector setup requires browser approval on every run, even when you supply an API key.

For Pi, Hermes, Conductor, or setup without connectors, run the Router installer:

curl -fsSL https://router.com/install.sh | sh

It opens your browser to sign in to Router and lets you choose your agents.

On a machine without a browser, create a key in API keys↗ and pass it in explicitly:

curl -fsSL https://router.com/install.sh | sh -s -- --api-key 'your-router-key'

If the CLI is already installed, run ramp router configure without --connectors for model-only setup.

To restore agents to their original model configuration, run ramp router unconfigure and select which agents you want to remove Router models from.

If you'd rather not pipe a script into your shell, you have a few options:

• Download the installer, read it, and then run it:

  Copy

  curl -fsSL https://agents.ramp.com/install.sh -o install.sh
  less install.sh
  sh install.sh --no-skills

  Open a new terminal if the installer updates your PATH, then run:

  Copy

  ramp router configure --connectors

• Install the Ramp CLI with Homebrew, then configure models and connectors:

  Copy

  brew install ramp-public/ramp/ramp-cli
  ramp router configure --connectors

• Skip the CLI entirely and set up your agent by hand. Each guide shows exactly which files to edit and what to put in them, so you can see every change before you make it.

Alongside adding the models that Router supports to your coding agent, Router also adds status information that shows which models served your session, whether Switchyard routing is enabled, and how much you saved by using Router.

This is currently supported for Claude Code, Pi, and OpenCode.

For Claude Code, we install a Claude Code status line↗ that shows which models served your session, how many requests each handled, the direct cost, and how much you saved by using Router:

Switchyard enabled · Claude Opus 5.5 ×12, Claude Sonnet 5.5 ×30
██████████░░░░░░░░░░░░░░ $4.09 vs $10.02 direct · 59% saved

The full bar reflects the full price of the session, and the filled part reflects how much you actually paid using Router. When the two prices match, the bar is hidden and the status line shows a single session total.

For Pi and OpenCode, we include a status line that shows what model served your last prompt, along with any savings from Router:

Switchyard enabled
Routed to: gpt-6.1-sol via openai  -50% vs Gpt 6.1 Sol
Ramp        █████████████░░░░░░░░░░░ $0.35
Gpt 6.1 Sol ████████████████████████ $0.71

The full bar is the direct price, and the filled part is the Router price. When Switchyard routing↗ is enabled, the first line says so, and the empty part of the bar is what routing saved against your selected model. When the two prices match, the bar is hidden and the status line shows a single session total.

Router's model catalog is regularly updated as new models are released. Most changes will be picked up by your coding agent, but support varies from agent to agent:

Every agent below can be set up by editing its configuration yourself, with nothing to install. Each guide lists the exact settings, explains what they do, and shows how to undo them.

Note that manual setup does not include all the functionality that the Ramp CLI offers. See the guides for each agent for more details.

For an agent that isn't listed, you can configure Router within it manually. Here is the key information you'll need:

• The base URL for Router is https://api.router.com/v1.
• You will need to create an API key on the API keys↗ page.
• Prefer the OpenAI Responses API; it supports Router's fallback lists and Router-only fields. If your agent only speaks the legacy OpenAI Chat Completions API, Router also serves POST /v1/chat/completions on the same models, with the limits described in Responses and Messages↗.
• If your agent supports model auto-discovery, Router provides an OpenAI-compatible listing API at https://api.router.com/v1/models.
• You will need to configure all the models you want to use with Router in your agent manually. Refer to Supported models↗ for a list of supported models.

Claude.ai Connectors are not supported with Router because they require an Anthropic subscription. Use Router connectors↗ for supported apps, or configure an equivalent MCP server directly in your coding agent.