How to connect FreeScout to OpenAI
Works with OpenAI
To connect FreeScout to OpenAI, open a mailbox's ReplyRabbit settings, choose OpenAI as the provider, paste an API key from your OpenAI account, pick a model, and save. The next customer email in that mailbox gets an OpenAI-written draft. OpenAI is the one provider available on the free plan, so this is the fastest way to see ReplyRabbit working before you decide on anything else.
Before you start
- ReplyRabbit installed and activated in FreeScout. If not, see How to set up ReplyRabbit for the full path.
- An OpenAI account with API access and billing set up. ReplyRabbit uses the API, not a ChatGPT subscription, and OpenAI bills that account directly for usage.
- An API key created in the OpenAI developer console. Copy it somewhere safe; you'll paste it once.
Connect OpenAI to a mailbox
Open the mailbox's ReplyRabbit settings
In FreeScout, open the mailbox, go to Settings, and choose ReplyRabbit. Settings are per mailbox, so repeat this for each inbox that should draft with OpenAI.Select OpenAI as the provider
In the AI Provider card, set Provider to OpenAI.Paste your API key
Paste the key into API Key. After saving, ReplyRabbit shows it masked and never displays the full key again.Choose a model
Leave the latest option selected unless your team has standardised on a specific model. Use Custom only when you know the exact model ID OpenAI supports on your account.Set a monthly token budget (optional)
Enter a Monthly Token Budget if you want ReplyRabbit to stop generating once the mailbox reaches a token total for the month. It counts input and output tokens together.Save and draft a test reply
Save, open a real customer conversation, and generate a draft. Read it before you send it, and adjust company context if the voice is off.
What OpenAI powers once connected
On the free plan, OpenAI writes draft replies and the rewrite options (nicer, shorter, more direct), with spam and noise filtering and smart reply routing around them. On Pro and Team, the same key can also power Next Actions checklists, AI Signals, and embeddings for semantic search over your knowledge base and products. You can point those background tasks at a cheaper OpenAI model than your drafts use; see Per-task AI models.
ReplyRabbit sends OpenAI a safe prompt cache routing key with each direct request so repeated support prompts route consistently. There's no cache toggle to manage, and the token budget still counts every token regardless of caching.
What gets sent to OpenAI
For each enabled feature, ReplyRabbit sends the conversation text needed for that task plus your company context, response style and examples. It doesn't send your whole mailbox, and it doesn't send anything for conversations it filters out as spam or noise. If some mailboxes handle data that can't go to a cloud provider, give those mailboxes a local model instead; see Choosing an AI provider.
If something doesn't work
| Symptom | Likely cause |
|---|---|
| "Invalid API key" | The key was truncated when pasted, or it belongs to a different OpenAI organisation. Re-copy it. |
| Drafts worked, then stopped | The free plan's 10 drafts a month ran out, the token budget was hit, or the OpenAI account has no credit. |
| Model error | The custom model ID isn't available to your account. Switch back to the latest option. |
| Draft ignores your policies | Company context or response style is empty. Fill them in and regenerate. |
More cases are in Troubleshooting ReplyRabbit and the FAQ.
Official references
Provider names, model lists, pricing, privacy terms, and legal guidance change. Check the current source before setting a production policy.