Skip to content
Elitea AI Docs

AI Providers

AI Providers is where a project registers the models and storage it uses, and picks defaults for each category. Settings here apply at the project level for every member.

AI Providers configuration cards grouped by category

Categories

Section titled “Categories”

LLMs

Language models, with Default, High-tier, and Low-tier selections.

Embedding Models

Used for indexing and semantic search.

Image Generation

Powers the Image creation module in chat.

Speech Recognition (ASR)

Server-side voice input.

Text to Speech (TTS)

Server-side spoken playback.

Each category shows its configured models as cards with a provider icon, name, and a badge for Default, High-tier, or Low-tier. Select a card to edit it, if your role allows.

Add a model or credential

Section titled “Add a model or credential”

  1. Create AI Credentials first

    Choose a provider — OpenAI, Azure OpenAI, Anthropic, Amazon Bedrock, Google Vertex AI, Ollama, vLLM, or EPAM AI Dial — and provide its endpoint and key. One credential can back several models.

  2. Add the model

    Select +, choose the model category, and fill in a display name, the exact provider model identifier, and — for LLMs — context window, max output tokens, and whether it supports reasoning or vision.

  3. Set as default

    Use the category’s dropdowns to make the new configuration the project default, or the high-tier / low-tier choice.

Speech recognition and text-to-speech fallbacks

Section titled “Speech recognition and text-to-speech fallbacks”

If no ASR model is configured, voice input falls back to the browser’s Web Speech API; if no TTS model is configured, playback falls back to the browser’s SpeechSynthesis API. Elitea determines ASR mode from the model name — whisper or transcribe selects batch transcription; anything else uses realtime streaming.

Troubleshooting

Section titled “Troubleshooting”
No access to LLM / model not found
Confirm the model name matches the provider’s identifier exactly and that your credential has access to it.
Authentication failed (401/403)
Recheck the API key or token for typos or expiry, and confirm the endpoint URL is correct for your provider.
Voice falls back to the browser voice unexpectedly
Confirm a default ASR or TTS model is actually selected in this page, not just configured.