Skip to main content
Every engineer needs a Weave Router key (rk_…). Beyond that, what you connect depends on how Claude Code or Codex is signed in, not on the plan name. If the client is signed in with a Claude or ChatGPT account, the router reuses that login automatically. If it is signed in with an API key, the router does not use that key.
Short answer for Claude Enterprise: no Anthropic API key is needed. Sign in to Claude Code with your Enterprise claude.ai account, install the router, and Claude turns run on your Enterprise plan. Anthropic bills that usage to your Enterprise organization exactly as it would without the router.
Weave Router Max seats (not the same thing as Claude Max) are limited to open-source models on every request, so a Claude or ChatGPT login doesn’t unlock Claude or GPT models on a Max seat. Plan coverage works on a Weave Router Boost seat or a router key with no individual seat. See billing.

​
The three things you might connect

Your Claude or ChatGPT login is never entered in Weave for ordinary use. It stays in Claude Code or Codex, and the router forwards it with each request it serves on that plan.

​
Claude Code

If ANTHROPIC_API_KEY is set anywhere in your shell or Claude Code settings, Claude Code uses it instead of your claude.ai login, and you lose plan coverage. Run unset ANTHROPIC_API_KEY, remove it from your shell profile and Claude Code settings, and restart Claude Code. Run /status in Claude Code to see which login is active.

​
Claude Enterprise and Team admins

  • Members sign in with the organization account they were invited with. Nothing is entered in Weave.
  • If you deploy managed settings with allowedProviders, add "customEndpoint" and set ANTHROPIC_BASE_URL to https://router.workweave.ai in the same managed settings file. Otherwise Claude Code refuses to send requests to the router.
  • forceLoginMethod and forceLoginOrgUUID keep working. The router key travels in its own header and does not count as a login.
  • Remote Control needs a direct connection to Anthropic, so it is unavailable while routing is on. See Remote Control troubleshooting.

​
Codex

A ChatGPT login covers OpenAI’s native Codex models only. The router checks how much included usage is left before it sends a turn to your plan. When that usage runs out, or the router picks a model outside the Codex family, the turn uses router credits or your own provider key if paid fallback is on, and stops with a limit error if it is off. The hosted router does not spend extra ChatGPT credits you bought on top of a Plus or Pro plan.

​
ChatGPT Business, Enterprise, and Edu admins

  • In workspace settings, turn on Allow members to use Codex locally (or Allow members to use Codex and Work locally).
  • If you pin Codex with managed configuration, keep forced_login_method = "chatgpt". Setting it to "api" makes every engineer an API-key user, so nothing is covered by your plan.
  • forced_chatgpt_workspace_id keeps working with the router.

​
When you need credits or your own key

Plan coverage only applies to the plan’s own model family: Claude models for a Claude plan, native Codex models for a ChatGPT plan. You need router credits or your own provider key when:
  • The router picks a model from another family, for example an OpenAI or open-weight model during a Claude Code session.
  • An engineer reaches their plan’s limit and paid fallback is on.
  • An engineer is signed in with an API key instead of a Claude or ChatGPT account.
Provider keys you add in Weave bill your provider account directly, and Weave charges a 5% platform fee against your credit balance, so a small funded balance is still needed. See billing. To keep everything on your plan, set subscription routing to Let router choose or Strict pass-through until low (Use prepaid credits only ignores plans entirely), restrict models to your plan’s family, and turn paid fallback off. Engineers then get a limit error instead of paid usage when their plan runs out.

​
Checking that it worked

Run a short prompt, then open router.workweave.ai. Turns served on a plan show $0 in credit spend while still appearing in usage reports. If plan-covered turns are spending credits, check for ANTHROPIC_API_KEY or an API-key Codex login first.
Last modified on October 27, 2026