Best OpenAI settings for ReplyRabbit
The best OpenAI settings for ReplyRabbit are mostly not about OpenAI. Leave the model on the latest option, then spend your effort on the fields that shape the draft: company context, response style and examples. Add a monthly token budget so costs can't surprise you, and on paid plans point background tasks at a cheaper model than drafts. This page goes through each setting in the order it pays off.
Model: start with the default
In the AI Provider card, the model dropdown offers the latest option, listed presets, and Custom. Start with the latest option. It's the one ReplyRabbit's prompts are tuned against, and it saves you from tracking OpenAI's model list.
Switch to a preset or custom ID only when you have a reason you can state: a cost target, a compliance requirement, or a measured quality difference on your own tickets. Model names and prices change often, so check OpenAI's current list before hard-coding one.
Company context: the setting that changes drafts most
The Response Style card holds the text that goes into every prompt. This is where most of the quality comes from.
- Company Name, Your Name and Your Role set who the draft is from.
- Business Website can auto-fill the fields below it: enter the URL and click Fetch.
- Company Context is plain factual background: what you sell, who your customers are, your return window, your support hours. Write it like a briefing for a new hire.
- Response Style describes the voice and the rules. "Warm but brief. Never promise a delivery date we haven't confirmed. Always link the relevant help article." That kind of thing.
- Examples are two or three real replies you'd be happy to send. They set the register without being copied.
A draft with good context and a modest model beats a draft with a top model and empty fields.
Token budget: a hard stop you'll be glad of
Monthly Token Budget stops ReplyRabbit generating once the mailbox reaches a token total for the month. It counts input and output tokens together, and it doesn't discount cached input even when OpenAI bills it at a lower rate, so it's conservative on purpose. Set it slightly above what a normal month uses, then raise it once you know the pattern.
ReplyRabbit sends OpenAI a safe prompt cache routing key automatically. There's no cache setting to tune.
Per-task models: cheaper where it's safe
On Pro and Team, the AI Provider section has separate subsections for Drafts, Embeddings, Mailbox previews and Attachment vision. Keep drafts on your best model. Leave previews and vision on Cheap default (recommended); they run far more often and a summary doesn't need your top model. Leave the extra API key fields blank so they reuse the key from Drafts. See Per-task AI models.
Auto Draft and Auto Actions
Auto Draft (New Email) writes a draft as each new customer message arrives, so agents open tickets with a reply waiting. Auto Actions does the same for the Next Actions checklist on paid plans. Both increase OpenAI usage in proportion to ticket volume, which is where the token budget earns its keep. Neither sends anything; drafts wait for an agent.
A settings checklist
- Provider OpenAI, model on the latest option.
- Company Context, Response Style and at least two Examples filled in.
- Monthly Token Budget set.
- Previews and vision on the cheap default (Team).
- Auto Draft on once the drafts read right.
The wider setup is in How to set up ReplyRabbit. If drafts still miss the mark, Troubleshooting ReplyRabbit covers the usual causes, and Choosing an AI provider explains when another provider might suit better. The FAQ answers the billing questions.
Official references
Provider names, model lists, pricing, privacy terms, and legal guidance change. Check the current source before setting a production policy.