OpenWhispr Proxy.
Connect OpenWhispr to Vercel AI Gateway and track your dictation usage.
Overview
OpenWhispr Proxy lets OpenWhispr use Vercel AI Gateway for audio transcription and transcript cleanup. It runs locally between the application and the gateway, adapting the audio API format and recording usage for both audio and text requests.
How it works
Run the proxy and configure OpenWhispr to send requests to it. The proxy maps the incoming parameters to the gateway's API and returns responses in the format OpenWhispr expects. A prebuilt Docker image is available through GitHub Container Registry, so you can run it without building the image yourself.
Each request is recorded in SQLite. The dashboard shows recorded costs, audio duration, activity over time, and usage by model. A searchable request table supports filters such as date range, operation, model, and cost, so you can inspect transcription and text-cleanup activity in one place.
Open `/dashboard` on your local proxy address to explore the dashboard.
Technical decisions
Adapt the existing client requests
I inspected the endpoints, parameters, and response formats OpenWhispr uses and implemented the corresponding routes in NestJS with the Vercel AI SDK. Text requests already work with the gateway's OpenAI-compatible API; the audio transcription format is the compatibility gap the proxy handles.
Keep request history locally
I added SQLite to retain per-request usage and cost records, then built a dashboard over that history. Docker volume persistence keeps the records between container restarts.