AI models

Every AI feature (editing code with AI, generating a domain, Maestro, the AI draft of the new project wizard) runs on the AIs of the organization. Each AI is a provider, a model and the credentials to reach it. You bring your own: requests go to your account at the provider.

An organization can define several AIs, for example Claude on the Anthropic API and GPT on Azure OpenAI. Users pick one per request in the AI selector, preselected on the default.

Adding an AI

Open Organization settings from the organization menu, tab AI. The Artificial intelligences list is visible to every member of the organization; only an Owner or an Admin can add, edit or delete an AI.

Click Add AI and fill in:

Field	Notes
Name	Unique in the organization, for example Claude for the teams
Provider	See the table below
Provider settings	The credentials and location the provider needs
Model / Deployment	Which model the requests go to
Maximum effort	Optional. The highest reasoning effort users can pick with this AI (Low, Medium, High, Extra high, Max). No limit by default
Default AI	Preselected in every project that can use it. Only one AI of the organization is the default

Providers

Provider	Settings	Model
Claude (Anthropic)	API key	Model, optional: claude-opus-5-5 when left empty
Claude (Azure AI Foundry)	Resource (the resource name, not a URL) and API key	Model: the name of the Claude deployment, optional: claude-opus-5-5 when left empty
Claude (AWS Bedrock)	Region, Access key ID and Secret access key	Model, required: the Bedrock model id or an inference profile. The key needs permission to invoke it
Gemini (Google)	API key	Model, required: use a model with function calling
GPT (Azure OpenAI)	Endpoint, API key and Reasoning model	Deployment, required: the name of the GPT deployment

Notes:

• The Azure OpenAI Endpoint must be an Azure HTTPS endpoint, like https://my-resource.openai.azure.com/. Turn on Reasoning model when the deployment is a reasoning model (for example GPT-5 or the o-series): it then accepts an effort level.
• Reasoning effort is offered on Claude with all five levels, and on GPT with Reasoning model on with Low, Medium and High. Without reasoning, and on Gemini, the choice is hidden.
• On Claude, the effort levels need Claude Opus 4.7 / Sonnet 5 or later. The default model is one.
• The prompts and tools of the agent were written and tested with Claude: the quality of the code agent depends on the model.

Credentials

The credentials are never shown again after saving: the form keeps a hidden value, and leaving it as it is keeps the stored key. A stored key is kept only while the provider, the resource, the endpoint and the access key stay the same; point the AI elsewhere and you type the key again. The stored keys are never sent to another provider.

On a self-hosted installation, set SECRETS_ENCRYPTION_KEY to have them encrypted at rest with AES-256-GCM. See Encryption at rest.

Deleting an AI

Deleting an AI also removes it from every project that uses it. A project left with no AI has the AI features turned off.

Which AIs a project can use

Below the list, the Project access table shows, for each project of the organization, the AIs it can use. Owners and admins edit a row and choose:

• All the AIs of the organization (default), including those added later; shown as All (default);
• Only the selected AIs: an AI added later is not included.

Selecting none turns the AI features off for that project, shown as AI turned off: Maestro is hidden, the AI endpoints answer that AI is not enabled, and nothing is sent to any provider.

The AI preselected in a project is the organization's default when the project may use it, otherwise the first one available.

The system AI (self-hosting)

A self-hosted installation can offer one more AI to every organization from its environment variables. It is listed as System AI, is read-only in the console (Configured in the installation's environment: it cannot be edited here.), and can be excluded from a project like any other.

AI_PROVIDER	Required variables	Optional
anthropic	ANTHROPIC_API_KEY	ANTHROPIC_MODEL (claude-opus-5-5)
azure-foundry	AZURE_FOUNDRY_RESOURCE, AZURE_FOUNDRY_API_KEY	AZURE_FOUNDRY_MODEL (claude-opus-5-5)
aws-bedrock	AWS_BEDROCK_REGION, AWS_BEDROCK_ACCESS_KEY_ID, AWS_BEDROCK_SECRET_ACCESS_KEY, AWS_BEDROCK_MODEL
gemini	GEMINI_API_KEY, GEMINI_MODEL
azure-openai	AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_API_KEY, AZURE_OPENAI_DEPLOYMENT	AZURE_OPENAI_REASONING (false)

• AI_MAX_EFFORT (low, medium, high, xhigh or max) caps the effort of the system AI. On an installation open to sign-ups it bounds the cost of a single question.
• AI_PROVIDER left empty with ANTHROPIC_API_KEY set means Claude through the Anthropic API.
• A provider named without one of its required variables stops the backend at boot, with the missing variables in the error.
• For Azure sovereign clouds, add the endpoint suffixes the installation accepts with AZURE_OPENAI_ENDPOINT_SUFFIXES, comma separated (for example .openai.azure.us,.openai.azure.cn).

With none of these variables set there is no system AI. See also Environment variables.