Connect a model provider

Chats and agents need a language model. A model provider is one connection to a provider: the model it should use, where the endpoint lives, and the credentials that authorize it. A local runtime starts with the curated model catalog but no connections, so the first chat asks you to add one.

Providers are user-level, not project-level. A connection you add is available to every project you can open.

Supported providers

Provider	Connection type	Credentials	Notes
LM Studio	Local, OpenAI-compatible	Optional API key	Default endpoint http://localhost:1234/v1
Ollama	Local, OpenAI-compatible	Optional API key	Default endpoint http://localhost:11434/v1
Ollama Cloud	Cloud, native	API key	Default endpoint https://ollama.com
OpenAI	Cloud, native	API key	Endpoint can be overridden
Anthropic	Cloud, native	API key	Claude models
Google Gemini	Cloud, native	API key	Gemini models
Google Vertex AI	Cloud, native	Application Default Credentials or API key	Also needs a Google Cloud project and location
DeepSeek	Cloud, OpenAI-compatible	API key	Default endpoint https://api.deepseek.com/v1
Mistral	Cloud, OpenAI-compatible	API key	Default endpoint https://api.mistral.ai/v1
OpenRouter	Cloud, OpenAI-compatible	API key	Default endpoint https://openrouter.ai/api/v1
Azure OpenAI	Cloud, native	API key	Needs the resource endpoint and deployment name
AWS Bedrock	Cloud, native	AWS access key and secret	Needs a region; optional session token
OpenAI-compatible	Any endpoint	Optional API key	Use for a gateway or a provider without a dedicated type

For each provider you choose a curated model from the catalog, or type any model id the provider supports. The connection also gets a display name, so a project can tell two connections to the same provider apart.

Add or edit a provider

Open LLM Providers from the user console and select Add Provider. Choose the provider type, pick or type the model, give the connection a name, then fill in the endpoint or credentials that provider requires. Enabled makes the connection usable; Default Provider makes it the first choice for new chats. Add Provider validates the connection before saving it. Edit reopens the same form for a saved connection, and missing required fields are reported in place instead of saving a partial connection.

1. LLM Providers groups connections by provider type. Select a provider tab to see its connections; each card shows the model, whether it is the default, and which credentials are configured.
2. Add Provider chooses a provider type, a curated model, a display name, and the credentials or endpoint it needs. Add Provider in the dialog validates the connection before it becomes available.
3. Edit reopens a saved connection with its model, endpoint, and availability. Stored credentials are not shown; leave the field blank to keep them.
4. Saving without a model or display name reports the missing fields in place, so nothing is stored until the connection is valid.

Manage connections

Each provider tab groups its connections and shows how many exist. A card reports the model, the provider type, the runtime model name, which credentials are configured, and a readiness state:

• Ready means the last validation succeeded.
• Needs attention means validation failed; open Edit and check the credentials or endpoint.
• Unknown means the connection has not been validated yet.

Edit changes the model, name, endpoint, or credentials. Leaving a credential field blank while editing keeps the stored value. Delete removes the connection without deleting the model catalog.

Credentials are stored by the local server and are never returned to the browser; the card only reports which fields are set.