AI provider configuration
Connect text and embedding models to a self-hosted Talqo deployment.
Talqo uses operator-owned provider credentials. Users with the ai_provider:manage permission configure one text model and one embedding model from Dashboard → AI configuration.
Providers
Talqo supports OpenAI, Anthropic, Google Gemini, Mistral, Azure OpenAI, Amazon Bedrock, and OpenAI-compatible endpoints. Azure OpenAI and Amazon Bedrock accept static credentials or the API process's deployment identity. Anthropic provides text models only, so its text configuration requires a separate embedding provider.
When both roles use the same provider account, the embeddings section shares the connection from Text generation by default, so Talqo stores only one encrypted credential set. Switch to Different provider to configure a separate account or provider for embeddings.
Model selection
After you complete a provider's connection settings, Talqo automatically loads the available model identifiers and offers them as suggestions in the model field. Providers return identifiers without classifying them as text or embedding models, so select the correct type for each role. If the list cannot be loaded or the provider does not support discovery, enter the exact model identifier manually.
OpenAI-compatible endpoints
OpenAI-compatible configuration accepts public, private, loopback, HTTP, and HTTPS base URLs for self-hosted inference. A user with ai_provider:manage can therefore make the API process contact internal network addresses. Grant this permission only to trusted operators.
Talqo sends credentials only to the configured origin, rejects non-HTTP protocols, does not forward credentials across cross-origin redirects, limits request duration and response size, and hides provider response bodies from dashboard errors.