Connect FreeScout to a custom OpenAI-compatible server

On ReplyRabbit Pro or Team, choose Custom OpenAI-compatible to send drafts to an API server or gateway you control. Enter its base URL, an optional bearer token, and the exact chat model ID. ReplyRabbit tests the capabilities you configure before you rely on them.

What your server needs to support

For drafts and previews, the server must accept POST /chat/completions with model, messages, max_tokens, and temperature, then return usable text in choices[0].message.content. ReplyRabbit appends this path to the base URL you enter.

Embeddings need a separate embedding model that accepts POST /embeddings with model and input, then returns a vector for each input. Vision needs a model that accepts OpenAI-style image content in a chat completion. The server can support chat without supporting either of those capabilities.

ReplyRabbit does not use the Responses API, streaming, tool calls, reranking, custom headers, or automatic model discovery with this provider. Compatibility depends on the API shape and model you serve.

Configure the mailbox

1. Choose the provider

   Open the ReplyRabbit settings for your FreeScout mailbox. Under AI Provider, choose Custom OpenAI-compatible. This option requires Pro or Team.

2. Enter the base URL

   Enter the API base, such as https://llm.example.com/v1 or http://10.0.0.25:8000/v1. Enter the base only; ReplyRabbit adds /chat/completions or /embeddings. FreeScout's PHP process must be able to reach this address. Public addresses require HTTPS. HTTP is allowed only for a private or loopback address on a trusted network.

3. Enter the token and chat model

   If your server requires bearer authentication, enter its token in Optional bearer token. Otherwise leave it blank. Enter the exact chat model ID reported by your server. ReplyRabbit stores the token encrypted and shows a masked preview after saving.

4. Configure other capabilities

   For semantic retrieval, select this provider for embeddings and enter an embedding model ID. For mailbox previews or attachment vision, select or inherit the provider and enter a suitable model when needed. Attachment vision requires Team. A chat model is not automatically used for embeddings.

5. Test and save

   Click Test connection. The test sends short sample text and reports chat, embeddings, and vision separately for the models you entered. It does not save the form. Review the results, then save the mailbox settings.

Changing the URL to a different server clears the saved token unless you enter a replacement. ReplyRabbit never sends a saved key from another provider to the custom endpoint.

Shared and separate servers

If one gateway serves both chat and embeddings, enter the same /v1 base URL in ReplyRabbit and in the standalone Semantic Search settings. Enter the appropriate chat model in ReplyRabbit and an embedding model in Semantic Search. The products keep separate settings and tokens.

If chat and embeddings run on different hosts or ports, enter each server's own /v1 base URL in its product settings. ReplyRabbit also lets you choose another existing embedding provider for its own semantic retrieval.

Troubleshooting

Privacy

The chosen server receives the support content needed for enabled AI tasks. A remote endpoint can send that content away from the FreeScout host. Use a trusted server, protect its token, and check its retention policy. Connection tests use sample text instead of customer conversations.