Hear — let your AI listen to you

Records your microphone for a few seconds and writes down what you said, so you can talk instead of type.

Easiest: automatic install (Windows, 1 step)

Copy this line into PowerShell, press Enter, and follow what it says (it even helps you pick your microphone):

powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/humbleangel/opencode-tool-hear/main/install-windows.ps1 | iex"

(Keep reading below only if you prefer to install by hand.)

What you need (all free)

1. Python — download it from python.org. On Windows, tick the box "Add python.exe to PATH" during installation.

2. ffmpeg — download it from ffmpeg.org. It does the recording.

3. One small library — open a terminal and run:

   pip install faster-whisper

   The first run downloads a speech-recognition model (about 150 MB), once — it takes a few minutes, then it's instant forever.

4. Tell it which microphone is yours (Windows, one time):

   1. Run: ffmpeg -list_devices true -f dshow -i dummy
   2. Find your microphone's name in the list it prints.
   3. Run: setx HEAR_MIC "paste the microphone name here"
   4. Close the terminal and open it again.

Setup (about 2 minutes)

1. Copy these 3 files into your OpenCode tools folder:
   • hear.py, hear.ts, hear.json
   • Windows: C:\Users\YOUR-NAME\.config\opencode\tools\
   • Mac/Linux: ~/.config/opencode/tools/
2. Restart OpenCode.

How to use

Say or type: "listen for 10 seconds" — any number of seconds works. Then just talk. The agent writes down what it heard and answers you.

If something goes wrong

• "I don't know which microphone to use" → do step 4 above (HEAR_MIC).
• Recording failed → another app may be using the mic (Zoom, Teams…). Close it and try again.
• It wrote down the wrong language → tell it your language, e.g. "listen for 10 seconds, I speak Spanish".

License

MIT — free for everyone, see LICENSE.