Bring your own model keys

You can point emma at your own model-provider accounts instead of using emma's. When you add your own key for a provider, emma uses it for that provider's models — you pay the provider directly with no emma markup, and that provider's models become available to your team.

In a nutshell:

• Keys are company-scoped and managed by an administrator
• A your-own key replaces emma's for that provider and unlocks its models
• You can also point emma at a self-hosted, OpenAI-compatible endpoint
• Which models are enabled is controlled per user

1. Add a provider key (admin)

An administrator adds keys under the company's model settings. Keys are stored encrypted and used only to call that provider on your behalf.

2. Point at a self-hosted endpoint (optional)

If you run your own OpenAI-compatible endpoint, add it as a base URL plus model id (and an optional key). emma reaches it behind SSRF protections — it must be an https URL and cannot target private or metadata-range addresses. For running models on infrastructure emma provisions, see Self-hosted inference.

3. Enable models for your team

Once a key is in place, its models can be enabled per user. Each person turns on the models they want in their own settings, then selects them in the model picker.

What's next

• Choosing models — select the models you've unlocked
• Self-hosted inference — run models on your own infrastructure
• The copilot — what the models power