KindPath Compass

A practice aid for compassionate listening in the presence of noise.

Three moves: texture recognition, emotional architecture, ZPB reimagining.

"May I join you for a while on your journey."

---

What It Does

KindPath Compass takes a text transcript, uploaded audio, or YouTube URL and runs three structured analysis prompts against it to surface:

1. Texture Recognition — what is actually present beneath the surface presentation
2. Emotional Architecture — what structures hold the situation in place
3. ZPB Reimagining — what genuine care as governing logic would produce

The result is a KindPath Compass reading: not a recommendation, a diagnosis, or a plan — a clearer view that the practitioner uses to inform their own judgement.

---

Install

# Clone and enter
cd kindpath-compass

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate       # macOS / Linux
# venv\Scripts\activate        # Windows

# Install dependencies
pip install -r requirements.txt

# Optional: audio/YouTube support (Whisper + yt-dlp)
pip install openai-whisper yt-dlp

Requirements

• Python 3.10+
• fastapi, uvicorn — API server
• httpx — HTTP client (LLM backend calls)
• python-multipart — file upload support
• openai-whisper (optional) — audio transcription
• yt-dlp (optional) — YouTube audio download

LLM Backend

The app will use Claude if ANTHROPIC_API_KEY is set in your environment, otherwise falls back to Ollama at http://localhost:11434.

# For Claude (recommended)
export ANTHROPIC_API_KEY=sk-ant-...

# For Ollama (offline)
brew install ollama && brew services start ollama
ollama pull llama3.2

---

Run

source venv/bin/activate
uvicorn app:app --reload --port 8008

Open http://localhost:8008 in a browser.

---

API Endpoints

Method	Path	Input	Description
GET	/	—	HTML UI
POST	/analyse	{"text": "..."}	Analyse text transcript
POST	/analyse/audio	multipart file	Transcribe then analyse
POST	/analyse/youtube	{"url": "..."}	Download, transcribe, analyse
GET	/health	—	Health check

Example: Text analysis

curl -X POST http://localhost:8008/analyse \
  -H "Content-Type: application/json" \
  -d '{"text": "He came in and sat down and immediately said he was fine."}'

Response includes texture, architecture, zpb_reimagining, and compass_reading fields.

---

Prompt Templates

Prompts are in prompts/:

• texture_recognition.md — Move 1 prompt
• emotional_architecture.md — Move 2 prompt
• zpb_reimagining.md — Move 3 prompt

These can be edited to adapt the Compass to different practice contexts.

---

Worked Example

See docs/example_mens_shed.md for a full worked example: a Men's Shed bereavement scenario run through all three Compass moves.

---

Methodology

The KindPath Compass methodology is documented in kindpath-canon.

The three moves derive from the practice framework: zones of proximity and belonging (ZPB), texture recognition as the entry point for authentic assessment, and emotional architecture as the structural layer beneath presenting behaviour.

---

KindPath Collective — built for practice, not for product.