---
title: "Per-agent provider overrides"
description: "Mix providers within a single workflow, different agents on different providers, with the resolution chain explained."
slug: guides/providers/per-agent-overrides
section: guides
subsection: providers
canonical_url: https://cerevisor.com/docs/guides/providers/per-agent-overrides
last_verified: 2026-05-18
last_verified_version: "1.2.0"
updated_at: 2026-05-18T15:08:18.053416+00:00
---

Every agent in a workflow resolves its provider via a four-level chain:

1. **Agent override** (`agent.providerId`): highest priority.
2. **Workflow override** (`workflow.settings.providerId`).
3. **Library default** (the entry marked as default).
4. **First enabled credential**: fallback if nothing else is set.

This means you can mix providers within one workflow without much ceremony.

## Three ways to set an override

### 1. The brain medallion on the agent card

Every agent has a small circular medallion at the bottom-right of its portrait: the "brain" badge. Its gradient color = the resolved provider's accent.

- **Click** to open the Provider & Model picker for this agent.
- **Right-click** to clear the override (revert to inherited).
- **Drag** a provider card from the Providers Dock onto the medallion to assign.

### 2. The Agent Config popup

Open Agent Config → **Provider** field. A chip-style picker shows all library providers plus an **Inherit** chip showing what the resolved provider would be without an override.

### 3. The Providers Dock

The dock lives in the right sidebar, below the Skills panel. Each library entry shows as a draggable gradient card. Drop on an agent's brain medallion to assign.

## Common mixed-provider patterns

### Senior + reviewer pair

- A research agent on Anthropic Opus (best synthesis).
- A reviewer agent on Anthropic Haiku (fast, cheap critique).

Saves money without losing the senior agent's quality on the main output.

### Sensitive data path

- Public-data agents on Anthropic.
- Sensitive-data agents on a local Ollama model (so the data never leaves your machine).

### Subscription-amortized writers

- Researcher and reviewer on Anthropic (pay per call).
- Writer on Codex CLI (free under your ChatGPT subscription).

Most expensive workflow component runs against the subscription you're already paying for.

### Cloud-resilient long runs

- Most agents on a fast cheap provider locally.
- The one agent that takes 20 minutes on Cursor cloud (survives a laptop sleep).

## Switching providers across the workflow

A workflow-level override (`workflow.settings.providerId`) sets the default for *every* agent in the workflow that doesn't have its own per-agent override. Useful when you want to try the same workflow against a different provider, e.g. swap from Anthropic to OpenAI for a budget test.

Set it in **Settings → Workflow** → **Provider override** chip picker.

## What happens to skills when you mix providers

Cerevisor's **skill-preflight** gate runs per agent. If you assign a skill that isn't compatible with the provider that agent resolves to, preflight blocks the run with a clear error. See [Skill compatibility with providers](../working-with-skills/skill-compatibility-with-providers.md).

This is the main reason per-agent overrides are useful: you can keep specialty skills on the provider that supports them while running the rest of the workflow elsewhere.

## What the Providers Dock looks like

A vertical strip of provider cards, each rendered as a gradient pill. The gradient is the provider's accent (Anthropic = warm orange; OpenAI-compatible = blue/purple gradient; Codex = green; Cursor = pink). One has a small "default" star icon.

Drag any card onto an agent's brain medallion to assign. A halo pulse animation confirms the change.

The dock is resizable, drag its top edge to grow it at the expense of the Skills panel above. See [Reference → Interactions](../../reference/interactions.md).