rye.ai
Menu
supervision docsProxy and CAv

Rye supervision workbench

Proxy and CA

Configure Rye's local proxy, install CA trust, and route model API traffic through the policy engine.

Start free trialView pricing

Start the proxy

Use:

exec:bash
rye up --install-ca --intercept-patterns openai.com,anthropic.com,claude.com,chatgpt.com,cursor.com,windsurf.com

The proxy becomes the control point for model API traffic. The CA installation allows Rye to inspect HTTPS model requests from supervised tools.

Routing scope

By default, use rye wrap or app shims to route an agent through the proxy. Rye injects HTTP_PROXY, HTTPS_PROXY, and NODE_EXTRA_CA_CERTS into the wrapped process. The installer and rye doctor --fix do not write global proxy variables into shell startup files.

Having no global shell proxy is expected when using shims. Explicit system proxy configuration through rye up --auto-configure-proxy or --auto-configure-proxy-pac is separate from wrapper routing.

If you intentionally want proxy variables persisted in your shell profile:

exec:bash
rye doctor --fix --set-global-http-proxy

This affects other commands in that shell, including curl, package managers, and cloud CLIs. Those commands can fail when the Rye proxy is stopped. Use wrapped sessions for agent-specific routing.

Intercept patterns

Intercept patterns select which requests Rye inspects after traffic reaches the proxy. They do not force a process to use the proxy. Keep the list narrow enough to avoid surprising unrelated traffic, but broad enough to include the model providers and agent vendors your team uses.

Common entries include:

exec:text
openai.com
anthropic.com
claude.com
chatgpt.com
cursor.com
windsurf.com

Verify routing

Run:

exec:bash
rye doctor
exec:bash
rye history --last 15m

rye doctor verifies local setup. rye history confirms that supervised requests are reaching Rye.

Stop the proxy

Use:

exec:bash
rye down

This stops the local proxy and restores proxy settings managed by Rye.

Common mistakes

  • Starting the agent before proxy variables are set.
  • Installing CA trust in one runtime but launching the agent from another.
  • Proxying the entire workstation during tests instead of using targeted intercept patterns.
  • Assuming the agent's own permission prompts are a substitute for network policy.