How Model Selection Works
Every message you send in Creor is routed to an LLM. Which model handles that message depends on a layered configuration system. Creor resolves the model in this order:
- Agent-level model -- if the active agent (build, plan, explore, etc.) has a model configured, that model is used.
- Global model -- the top-level "model" field in your creor.json or the model selected in the Settings UI.
- Creor Gateway default -- if no model is explicitly configured, the Creor Gateway selects a default model for you.
Models are referenced in the format provider/model-id, for example anthropic/claude-sonnet-4-20250514 or creor-gateway/google/gemini-2.5-pro.
Default Model
The default model is the model used for all agents unless overridden. Set it in your creor.json at the project root or in ~/.creor/creor.json for a global default:
You can also select the default model from the Settings UI inside Creor. Open the command palette and search for "Creor: Settings" or click the model name in the status bar.
Small Model
Creor uses a separate "small model" for lightweight tasks like generating session titles and summaries. This keeps costs low for background operations. Configure it with the small_model field:
Tip
small_model, Creor uses a cost-effective default from the gateway. You only need to override this if you want a specific model for background tasks.Agent-Specific Models
Each agent in Creor can use a different model. This is useful when you want a powerful reasoning model for planning and a fast model for building. Configure per-agent models in the agent section:
The agent-level model always takes priority over the global model. This lets you mix providers freely -- use Anthropic for building, Google for exploration, and OpenAI for planning, all in the same workspace.
Switching Models
Via Settings UI
Open the Creor settings panel (command palette or status bar), navigate to the Models section, and pick your preferred model from the dropdown. The change takes effect immediately for new messages.
Via creor.json
Edit the creor.json file in your project root. Creor watches this file and picks up changes automatically -- no restart required.
Filtering Providers
If you only want to use specific providers, use enabled_providers to whitelist them, or disabled_providers to blacklist specific ones:
Note
enabled_providers is set, only those providers are available. All others are ignored regardless of whether they have valid credentials.Supported Providers
Creor supports the following LLM providers. Each provider has its own authentication method and model catalog.
| Provider | Key Models | Auth Method | Best For |
|---|---|---|---|
| Creor Gateway | Claude, GPT, Gemini, Grok, and more | Creor account (built-in) | Getting started, no API key needed |
| Anthropic | Claude 4 Opus, Claude 4 Sonnet, Claude 3.5 Haiku | API key | Complex reasoning, large codebases |
| OpenAI | GPT-4o, o3, o3-mini, o4-mini, GPT-4 Turbo | API key | Fast responses, broad knowledge |
| Google AI Studio | Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 3 | API key | Long context, multimodal |
| Google Vertex AI | Gemini models (enterprise) | GCP credentials | Enterprise deployments |
| AWS Bedrock | Claude, Llama, Mistral, Nova | IAM credentials | AWS-native infrastructure |
| Azure OpenAI | GPT-4o, GPT-4 Turbo, o-series | API key + endpoint | Azure-native infrastructure |
| OpenRouter | 100+ models | API key | Multi-model access, one key |
| Groq | Llama, Mixtral, Gemma | API key | Ultra-fast inference |
| Together AI | Llama, Qwen, DeepSeek, Mixtral | API key | Open-source models |
| DeepInfra | Llama, Mistral, Qwen | API key | Cost-effective open models |
| Cerebras | Llama | API key | Wafer-scale inference speed |
| Mistral | Mistral Large, Codestral, Ministral | API key | European AI, code generation |
| Cohere | Command R, Command R+ | API key | RAG and enterprise |
| Perplexity | Sonar models | API key | Web-grounded answers |
| xAI | Grok 3, Grok 3 Mini | API key | Reasoning, real-time knowledge |
| Vercel | v0 models | API key | Vercel platform integration |
| GitHub Copilot | GPT-4o, Claude, Gemini via Copilot | GitHub auth | Existing Copilot subscriptions |
Provider Authentication
Most providers require an API key or credentials. You can set these up in three ways:
- Creor Settings UI -- navigate to the provider settings and enter your API key directly. Credentials are stored securely in the OS keychain.
- Environment variables -- set provider-specific env vars like ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.
- creor.json -- for advanced provider configuration (custom endpoints, options), not for storing API keys.
Warning
creor.json. Use the Settings UI or environment variables instead. The OS keychain provides encrypted storage for your credentials.