Managing the Provider Library

Adding, removing, renaming, setting defaults, and the deletion guard that prevents breaking active workflows.

Open Settings → Providers to manage the library. The tab lists every credential you've added.

Adding a provider

Click + Add provider. The Provider Setup wizard opens. Pick a type (Anthropic / OpenAI-compatible / Codex CLI / Cursor Agent) and walk through the type-specific setup.

See Connect a model provider for the per-type walkthroughs.

Renaming a provider

Click the pencil icon next to the provider's name. Type a new label. Press Enter.

Names are display-only; they don't affect behavior. Make them descriptive ("Ollama: M2 MacBook", "OpenRouter via personal account") if you have many.

Setting the default

Click Set as default on any entry. The old default loses the badge; the new entry gains it. Background features (chat builder fallback, retrospectives, etc.) switch over immediately.

You can't unset the default, there's always exactly one.

Removing a provider

Click the trash icon on the entry. Cerevisor checks if any open workflow has agents that reference this provider:

  • No agents reference it: Removal proceeds immediately.
  • One or more agents reference it: Removal is blocked. A toast lists the affected agents (workflow name + agent name).

To fix:

  1. Open each affected workflow.
  2. For each affected agent, change the provider (Agent Config → Provider) to a different library entry, or revert to inherit.
  3. Re-try the removal.

This guard exists because silently deleting a provider would leave agents in a broken state.

Key rotation

To rotate an API key:

  1. Open Settings → Providers.
  2. Click Edit on the entry.
  3. Paste the new key.
  4. Click Test connection.
  5. Save.

The OS keychain entry is overwritten. Workflows using this provider pick up the new key on their next run, no restart needed.

Multiple OpenAI-compatible entries

The openai-compatible type is the only one that allows multiple entries. Common pattern:

  • OpenAI (hosted): for production work where reliability matters.
  • Ollama (local): for cheap iteration during development.
  • OpenRouter: for trying exotic models.

Each entry is independent. The default is one of them; agents can override per-agent to any other.

Workflow-level provider override

In addition to the library default, each workflow can pin its own default in Settings → Workflow → Provider override. This is per-.cerevisor-file; it travels with the workflow when you share it.

If a recipient doesn't have a matching credential, Cerevisor falls back to the library default at run time.

Model defaults per provider

In each library entry's settings, the Default model picker lets you set the model that (auto) falls back to. Useful when you have a strong preference (e.g. "always Sonnet on this provider, never Opus").

The model dropdown is populated from /v1/models on the provider, refreshed every 5 minutes (cached locally for that window).

Auth troubleshooting

Some failure modes by provider:

Provider Symptom Likely fix
Anthropic 401 Key revoked or wrong. Regenerate in console.
OpenAI-compatible (OpenAI) 401 Same. Check platform.openai.com.
Ollama ECONNREFUSED Ollama isn't running. ollama serve or open the desktop app.
Codex CLI "Codex not found" Codex CLI not installed, or PATH issue. Reinstall: npm install -g @openai/codex.
Codex CLI "Not signed in" Re-run the device-auth step.
Cursor Agent "@cursor/sdk not installed" Install: npm install -g @cursor/sdk (or check the install command Cerevisor surfaces).
Cursor Agent "Invalid API key" Regenerate at cursor.com/dashboard.

Back to docs