Anthropic provider
Setting up and using Anthropic Claude (Opus, Sonnet, Haiku) in Cerevisor.
Setup
You need an API key from console.anthropic.com.
- Settings → Providers → + Add provider → Anthropic.
- Paste your API key into the API key field.
- Click Test connection. Cerevisor calls Anthropic's
/v1/modelsendpoint to confirm the key works. - Click Save.
The key is stored in your OS keychain. It never lives in plain text on disk or in any .cerevisor file.
Models
Cerevisor reads the live model list from /v1/models and surfaces it everywhere a model picker appears. The most-used Claude models you'll see:
| Model | Best for | Notes |
|---|---|---|
| Claude Opus 5 | Highest quality, deepest reasoning, long autonomous runs. | What the Opus tier setting now picks. Same price as Opus 4.8, so nothing gets more expensive by switching. Earlier Opus 4.x models stay available if you type their id. |
| Claude Sonnet 5 | Balanced — strong quality at lower cost. | What the Sonnet tier setting picks; the default for most workloads. |
| Claude Haiku 4.5 | Fastest, cheapest. | Great for reviewers, classifiers, simple transforms. |
The frontier-tier Claude Fable 5 is also available when your account has access to it; it costs more per token than Opus 5, so reserve it for the work that needs it. For models that support reasoning, the model picker shows a Thinking control (Off → Ultra) that sets how much the model may think before answering.
Cost
Cerevisor reads token usage from each response and multiplies by the per-model pricing table. The status bar shows running cost during a run; the audit log saves the breakdown.
Per-token prices update as Anthropic releases new models. Cerevisor's pricing table is hardcoded. If you see costs that look stale after a model launch, restart Cerevisor (the table is loaded at startup) or wait for the next release.
Default model
In the Library entry, you can set a default model for this provider. When an agent's model preference is (auto), it falls back to this default.
Headers and behavior
Cerevisor uses the official @anthropic-ai/sdk. Standard retry semantics, prompt caching (when supported), tool use, and vision are all wired through automatically.
Prompt caching is enabled by default when the API and the model support it; this can substantially reduce cost for workflows with shared context (e.g. multi-agent workflows reading the same project folder).
Common errors
| Error | What it means | Fix |
|---|---|---|
| 401 Unauthorized | Invalid API key. | Re-check the key. Generate a new one in the Anthropic console if needed. |
| 429 Rate limited | You hit your account's rate limit. | Wait and retry, or upgrade your Anthropic plan. |
| Network error | Anthropic is unreachable. | Check your network. Cerevisor retries automatically with backoff. |