> ## Documentation Index > Fetch the complete documentation index at: https://docs.sentivue.com/llms.txt > Use this file to discover all available pages before exploring further. # Connections > Store your provider credentials once and use them across every workflow A **connection** holds the credentials for one provider account. You add it once and every workflow in your organization can use it. Credentials live here and nowhere else. On [Model Configurations](/talk/configurations/inference-providers) you pick which model, voice, and language to use — not which key authenticates the call. ## What belongs in a connection Whatever identifies the account. Most providers ask only for an API key; some need one extra detail: | Provider | Also needs | | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------- | | [Azure OpenAI](/talk/configurations/providers/azure-openai) | Endpoint URL | | [Azure Speech](/talk/configurations/providers/azure-speech) | Region | | [Vertex AI Gemini](/talk/configurations/providers/vertex-ai), [Vertex AI Gemini Live](/talk/configurations/providers/vertex-ai-live), [Google Cloud](/talk/configurations/providers/google-cloud), [Gemini TTS](/talk/configurations/providers/google-gemini-tts) | Service-account JSON instead of an API key | | [MiniMax](/talk/configurations/providers/minimax) | Group ID, when the connection drives the voice | | [AWS Bedrock](/talk/configurations/providers/aws-bedrock) | Access key and secret key, no API key | ## One connection can drive several services Providers often cover more than one part of the pipeline. A single OpenAI connection can power the language model, the voice, and the transcriber — you do not add it three times. You can turn individual services off on a connection if you want it used for some but not others. ## Adding a connection 1. Go to **Connections** in your dashboard. 2. Choose a provider. 3. Fill in the fields and save. Some providers are checked against the vendor as you save, so a bad key is caught straight away. Others are only stored, and a wrong key first shows up on a live call — each provider's page says which. Place a test call after connecting either way. ## Connection states | State | Meaning | | --------- | ------------------------------------------------------------------- | | Connected | The provider accepted the credentials | | Failed | The provider rejected them. The error is shown | | Unknown | The provider could not be reached, so nothing was proven either way | You can re-test a connection at any time from its row. ## Staying healthy SentiVue Talk re-checks every connection daily and emails you when one starts failing, so an expired key surfaces before it takes down your calls. If a provider rejects a live call, the connection is marked failed straight away. You get **one** email about it, not one per affected call — and only after SentiVue Talk re-checks the credentials itself, since a provider having a bad hour should not send you an alert about a key that is fine. ## Multiple accounts with one provider You can add more than one connection per provider, each with its own label. Use this to separate environments, teams, or billing accounts. Renaming is safe. ## Removing a connection Delete it from its row. Move any workflow using it to another connection first, since a workflow left pointing at a deleted connection has no credentials to run with. ## Provider setup guides Providers that generate the agent's responses Providers that turn responses into speech Providers that turn speech into text Speech-to-speech providers that replace all three Providers that add video to a web call