GoalsGuild Coach Setup Tutorial
Follow this guide to configure BlackHole, complete app settings, and run Ollama with the best model for your machine.
On Windows? See the Windows setup tutorial (Voicemeeter + Ollama).
1) Install and configure BlackHole (macOS)
- Install BlackHole 2ch using Homebrew:
brew install --cask blackhole-2ch - Open Audio MIDI Setup and create an Aggregate Device with:
- BlackHole 2ch (system/interviewer audio)
- Built-in Microphone (your voice)
- Set your call app output to BlackHole and input to your microphone (or the aggregate device, depending on your routing preference).
If speaker labels are inverted in transcripts, enable Speaker Swap in settings.
2) Download and configure Ollama
- Install Ollama:
brew install ollama - Start Ollama:
ollama serve - Pull one model:
ollama pull mistral:7b-instruct
3) Suggested Ollama models by machine
| Machine profile | RAM (recommended) | Suggested model | Why |
|---|---|---|---|
| Base machine (older Intel / 8 GB) | 8-16 GB | phi3:mini or qwen2.5:3b | Fastest response with lower memory usage. |
| Mainstream (M1/M2 or modern i5/i7) | 16 GB | mistral:7b-instruct | Balanced quality and speed for live coaching. |
| Power user (M2 Pro/M3 Pro / high-end PC) | 24-32 GB | qwen2.5:14b (quantized) | Higher-quality reasoning with acceptable latency. |
| Workstation (M3 Max / 64+ GB) | 48-64+ GB | llama3.1:70b (quantized) | Best quality, suitable for deep post-call analysis. |
Start with the smallest model that keeps real-time suggestions responsive, then scale up only if you need higher answer quality.
How to identify your machine tier quickly
Use these quick checks and then match your machine to the table above.
- macOS: Apple menu -> About This Mac. Check chip (Intel/M1/M2/M3) and memory (8 GB, 16 GB, 24 GB, etc.).
- Windows: Settings -> System -> About. Check Installed RAM and processor generation.
- Rule of thumb: prioritize RAM first, then CPU/GPU tier. If unsure, choose the lower model class and move up only if latency stays acceptable.
4) Configure all key app settings
- LLM Provider: set to
ollama. - OLLAMA_BASE_URL:
https://www.GoalsGuild.com. - LLM Model: choose from the table above.
- Whisper Model: use
large-v3-turbo-q4on Apple Silicon. - Transcription Provider:
localfor offline mode. - Segment Duration: 2.5-3.5 seconds.
- Segment Overlap: 0.3-0.6 seconds.
- BlackHole Device Name:
BlackHole 2ch. - Aggregate Device Name: the exact name from Audio MIDI Setup.
- BlackHole Channel Count:
2. - Sample Rate:
16000(default) or44100. - Language Hint: set interview language (or auto).
- Suggestion Interval: 15-30 seconds depending on preference.
5) Validate your setup
- Open app Settings and click connection tests where available.
- Start a short mock call and verify transcript and live tips.
- If transcript speakers are swapped, toggle Invert Speakers.