Automatic fallback
If the pinned provider fails, the call moves to the next enabled provider instead of failing the email.
RelayRow does not lock your inbox to one AI vendor. It connects to OpenRouter for hundreds of models, or directly to Anthropic, OpenAI, Gemini, Groq, DeepSeek, Mistral, xAI, Together or any OpenAI-compatible API — and lets you pick a different model for each job, with automatic fallback and every call metered at its real token cost.
Updated
Each AI feature has its own routing: a pinned provider and model, temperature, token limit and fallback.
| Feature | What it does | Typical choice |
|---|---|---|
| Triage | Spam, category, urgency, sentiment, intent, summary | Fast, low-cost model |
| Replies | Drafts and auto-replies from your knowledge | Strongest model you trust |
| Rules | Yes/no questions inside automations | Fast model |
| Extraction | Pulls fields out of messages for rules | Fast model with structured output |
| Writing assistant | Improve, shorten, change tone, translate | Mid-size model |
| Summaries | Thread summary, key points, next step | Mid-size model |
| Knowledge embeddings | Retrieval over your knowledge base | An embedding model |
If the pinned provider fails, the call moves to the next enabled provider instead of failing the email.
Token cost is converted to credits per workspace wallet, with an append-only ledger.
Monthly caps, a minimum per call, and the option to block AI when a wallet is empty.
Every call is logged by feature, provider, model, mailbox and rule — exportable to CSV.
Provider keys are stored in Supabase Vault and never reach the browser.
Switch a feature to a newer model in AI Studio; nothing else changes.
| Aggregator | OpenRouter (hundreds of models, live pricing) |
|---|---|
| Direct | Anthropic, OpenAI, Google Gemini, Groq, DeepSeek, Mistral, xAI, Together AI |
| Custom | Any OpenAI-compatible endpoint |
Connect your first domain in minutes — no mail server, no IMAP, no glue code. AI and automations included.