How to connect FreeScout to Ollama
Works with Ollama
Connect FreeScout to Ollama by running Ollama on the same machine as FreeScout, pulling a model, and then choosing Ollama (Local) as the mailbox provider in ReplyRabbit with the default server URL http://localhost:11434. There's no API key. Drafts are generated on your hardware, so no conversation text leaves the server and there's no per-token bill. Local providers are part of ReplyRabbit Pro and Team.
What you need before FreeScout is involved
- A machine that the FreeScout PHP process can reach at
http://localhost:11434. In practice that means Ollama on the FreeScout host itself, unless your ReplyRabbit build documents another supported URL. - Enough RAM and, ideally, a GPU for the model you want. Small models run on a modest server; larger ones need real hardware.
- A model pulled into Ollama and tested from the command line, so you know it works before you blame ReplyRabbit.
Test Ollama on its own first:
ollama pull <model-name>
ollama run <model-name> "Write a two-sentence reply thanking a customer for their order."If that answers in a reasonable time, you're ready.
Connect Ollama in ReplyRabbit
Open the mailbox's ReplyRabbit settings
In FreeScout, open the mailbox, choose Settings, then ReplyRabbit. The mailbox must be on Pro or Team.Choose Ollama (Local)
In the AI Provider card, set Provider to Ollama (Local). The API key field isn't used.Confirm the server URL
Leave the URL ashttp://localhost:11434unless your installed version documents a different supported address.Enter the model name
Pick a model option or type the exact name of a model that's installed in Ollama. If the names don't match, generation fails.Save and time a draft
Save, open a customer conversation, and generate a draft. Note how long it takes; local latency is the thing most likely to surprise you.
What changes with a local model
- Privacy. Drafts, next actions and AI Signals are all generated locally. Nothing goes to a cloud provider for that mailbox.
- Cost. Ollama doesn't charge per token. ReplyRabbit still records usage so budgets and analytics work.
- Caching. Local models don't use provider prompt caching. There's no setting to change.
- Quality and speed. Both depend on the model and the hardware. Expect to try two or three models before settling.
You can mix. Keep a sensitive mailbox on Ollama and a general one on a cloud provider, because providers are chosen per mailbox.
Troubleshooting
| Symptom | What to check |
|---|---|
| Invalid Ollama URL | Use http://localhost:11434. Other hosts and ports aren't accepted unless your build documents them. |
| Model not found | The name in ReplyRabbit doesn't match ollama list exactly. |
| Draft times out | The model is too large for the machine, or Ollama isn't running. Try a smaller model. |
| Drafts are weak | Try a stronger local model, or move that mailbox back to a cloud provider. |
| Ollama option is disabled | The mailbox is on the free plan. |
For the LM Studio equivalent (default http://localhost:1234), and for the wider trade-offs, read Choosing an AI provider. The setup guide for the rest of the mailbox is How to set up ReplyRabbit, and general fixes are in Troubleshooting ReplyRabbit.
Official references
Provider names, model lists, pricing, privacy terms, and legal guidance change. Check the current source before setting a production policy.