Creating an LLM Provider
How to connect a cloud or self-hosted LLM backend to the platform.
An LLM provider is a connection to a language model backend. It stores the information the platform needs to authenticate with and send requests to that backend - nothing more. Providers don't have personalities or prompts of their own; that customization happens one layer up, at the agent level.
Supported provider types
The provider picker groups available backends into three categories:
Hosted APIs from major vendors:
OpenAIAnthropicGemini(Google)DeepSeek
No built-in model hosting
The platform itself does not run or host language models - it connects out to them. For self-hosted options, the provider's URL must point to an LLM server that's already running and reachable somewhere external to the platform (on your own infrastructure, a private network, or a cloud VM). If you need help standing up a self-hosted model server, our team can assist, and native self-hosted infrastructure is something we're actively exploring for a future release.
Creating a provider
Open the LLM Providers section
Navigate to Settings → AI → LLM Providers and select Add Provider.
Choose a provider type
Select the backend you want to connect from the grouped list. The fields shown next are the same for every type, but the values differ.
Fill in connection details
- Alias (required) - a friendly label to identify this provider throughout the platform (e.g.
OpenAI - Support Team). - Base URL (required) - the API endpoint of the provider:
- For cloud vendors, use their standard API endpoint (e.g.
https://api.openai.com/v1). - For self-hosted servers, the externally reachable address of your running model server (e.g.
https://models.example.comor an internal network address the platform can reach).
- For cloud vendors, use their standard API endpoint (e.g.
- API Key (optional) - your secret key from the provider's dashboard. Required for cloud vendors; only needed on self-hosted servers that require authentication.
Reachability
The Base URL must be reachable from the platform. localhost or loopback
addresses won't work unless the platform itself is running on that same
machine or network.
Save and verify
Click Add to save the provider. To confirm everything works, start creating an agent with this provider: if the platform can reach the endpoint, the model list will populate automatically. If it doesn't, double-check the Base URL and API key.
Using multiple providers for the same vendor
There's no restriction against creating several providers of the same type. A common pattern is registering two OpenAI providers with different API keys - for example, one per team or project - so that usage and cost can be monitored independently from OpenAI's own dashboard, rather than blending together under a single key. The same idea applies to any other vendor: each provider you create is a fully independent connection, even if they point at the same underlying service.