Connecting Claude

Every stage of the pipeline (implement, review, verify, security) is a call to the model provider you connect. You set this up in step 3 of the setup wizard, and can change it any time from Settings → Claude & models.

Providers

• Anthropic API key, from the Anthropic Console. This is what most installs use.
• Amazon Bedrock, Google Vertex AI, or Microsoft Foundry, for enterprise deployments that route model access through their existing cloud provider.

How billing actually works

Your key, your bill. Anthropic bills you directly for model usage. Reqursor never sees or resells tokens. Your code is sent only to the model provider you choose, as working context for the ticket being worked on; nothing about pricing or usage is proxied through Reqursor's own infrastructure.

Whichever provider you use, the key itself never reaches the agent process. A local gateway holds the real credential and injects it per request, so even a compromised or misbehaving agent run can't read it out of its own environment. See Agent isolation and secrets for the full picture.

Test connection

Paste your key and run Test connection before it's saved. It calls the provider's model list directly, so a bad key, an out-of-credit account, or a permission problem is caught here, not partway through your first ticket. A model your key doesn't have access to is flagged the same way, before anything tries to run against it.

Choosing models

Pick a default model and effort level, then override either per stage if you want implement to use a different model than review, for example. The packaged defaults:

New models Anthropic ships, such as a newer Opus or a lighter model for cheaper stages, appear in the picker automatically once your key has access to them; nothing to configure to pick them up.