Bring your own key
Use your own Anthropic, OpenAI, or xAI API key, so your provider bills the model directly, and see what stays on your Evergreen bill.
By default Evergreen pays the model provider and bills you for it at cost. If your company already has an Anthropic, OpenAI, or xAI account, a workspace Owner or Admin can set the workspace’s own key instead. Your provider then bills the model use directly. An xAI key runs Grok Build, which is in beta and turned on for each workspace.
Set a key
-
Open Inference
Go to Workspace settings › Inference. Evergreen pays for inference is the default and shows the providers Evergreen pays for. Your own keys lists the providers that run on a key of yours.
-
Pick the provider
Under Your own keys, choose Add a key and pick the provider. A provider that can’t take a key says why, such as On your key.
-
Paste the key
Enter it in API key and choose Save key. Evergreen checks it with the provider and stores it encrypted. After that you only see its last few characters.
The key applies to the whole workspace, from each session’s next turn. Its row under Your own keys shows the end of the key and who set it, with Replace to change it. To go back, choose Use Evergreen inference on the row and confirm. Evergreen deletes the stored key. The key still works at your provider until you revoke it there. If the page says a provider’s models run only on your own key, its row has Remove key… instead, and those models stop until someone adds a key again.
Only Owners and Admins can set keys, and not everyone’s workspace offers the option. If it isn’t offered, the page says so.
What your provider bills and what Evergreen bills
On a turn that runs on your key, your provider bills the model use and Evergreen books no inference charge. Everything else stays on your Evergreen bill:
- The platform fee.
- Agent time.
- Workspace storage.
- Search and knowledge, because embeddings and reranking can’t run on your key.
- Model calls Evergreen makes for your workspace that your key doesn’t serve. An Anthropic key also serves background work. OpenAI and xAI keys don’t. If the Anthropic row says Re-enter the key so background work runs on it too, choose Replace and paste the same key.
The Inference page shows an estimate of the model use on your key, at list price, marked as billed by your provider. It’s an estimate, and Evergreen doesn’t charge it. See Pricing for the rates.
Note
Model use on your own key still counts toward your budgets, at list price.
When your provider refuses the key
If your provider declines the key, for example because the account is out of credit or reached its spend limit, the session stops and says why. Your work is kept, and your Evergreen balance isn’t charged. The session names the provider’s account and offers the fix, such as adding credit there or replacing the key, and Retry session picks it back up. A person who can’t fix it can ask whoever set the key, or an admin, from the same place.
A Claude plan isn’t a key
Signing in with a Claude Pro, Max, or Team plan isn’t supported. An Anthropic API key is the way in. For how plans with API credits fit, read the answer in the FAQ.
Your free week and your own key
On your own key you don’t need a card for your free week. It starts when your first session runs on the key, and search is free during the week too. A workspace in its free week with its own key and no funds runs on that key only, so models your key doesn’t serve still need funds. When the week ends, the platform fee and agent time apply to turns on your key too. See Pricing.
Last updated