Vendor setup · Intelligence · Local
Set up LM Studio
LM Studio runs a language model on your own Mac and shapes your transcript into your template without sending it to any vendor. It is the only fully local note-shaping path in DirectScribe.
Intelligence (note-shaping) Fully local — no BAA needed No API cost
- Role in DirectScribe
- Intelligence — molds the transcript into your note template.
- Where the text goes
- Only to a loopback server on this same Mac. DirectScribe accepts only a loopback base URL (
127.0.0.1/localhost) and rejects any remote host — even over HTTPS — so the transcript never leaves this device. - Privacy paperwork
- None — there is no external vendor to sign an agreement with.
- Cost
- No API fee. You provide the hardware and electricity.
Remember what LM Studio does and does not cover: it makes note-shaping local. At launch, transcription is still a cloud step — you pair LM Studio with a cloud transcription vendor such as Deepgram.
Get it running
Install LM Studio
Download LM Studio for macOS from lmstudio.ai and install it.
Download a model
In LM Studio, search for and download an instruction-tuned model that fits your Mac's memory (a Gemma- or Llama-class instruct model is a good start).
Start the local server
Open LM Studio's Developer / Local Server tab, load your model, and start the server. It listens on
http://localhost:1234/v1by default.Point DirectScribe at it
Open Settings → Providers → Intelligence, choose LM Studio, and set the base URL to
http://localhost:1234/v1. You choose and load the model inside LM Studio — DirectScribe simply uses whatever your local server is serving, so you can leave its model field blank. The base URL must stay onlocalhost; DirectScribe rejects any non-loopback address.
A local-path trade-off, disclosed
Enable the bearer token
On macOS, a local server on localhost is not authenticated between processes — in principle another program on your Mac could bind the same port. DirectScribe accepts only a loopback base URL (rejecting every remote host, over HTTP or HTTPS) and checks the server's response shape, but we recommend turning on LM Studio's optional bearer token and entering it in DirectScribe's LM Studio settings for an extra layer. We disclose this rather than pretend the local path has no trade-offs.