Help that knows when to stop...
Most chatbots want you to keep talking. This one wants you to leave and go live your life.
empathySync.mp4
Every AI assistant is built to keep you talking. empathySync is the only one built to make itself less needed.
Full help for practical tasks. Deliberate restraint on personal ones. When it detects you're spiralling, it redirects you to real humans - not more conversation. Everything runs on your hardware via Ollama. No cloud. No data harvesting. No engagement optimization.
The things that matter most to you deserve a tool that knows the difference.
Why it's built this way
Every person should have the right to an AI system that is entirely their own. Not rented. Not monitored. Not optimised for someone else's engagement metrics. Yours - running on your hardware, answering only to you, storing nothing it doesn't need to.
For sensitive, personal things - how you're feeling, your relationships, your health, your money - you deserve something local, private, and restrained. For complex tasks that need serious compute, use the cloud AIs. That's a reasonable division.
But the part of AI that handles your personal life should belong to you.
This isn't a feature. It's the point.
If you're a developer who wants a privacy-respecting AI assistant with no API keys, no subscriptions, and no data leaving your hardware - this is for you.
If you're building ethical AI tooling and want a reference implementation that optimises for user autonomy rather than engagement, the architecture is fully documented and embeddable.
If you're a therapist, counsellor, or domain expert who wants to shape how an AI responds to emotional content - see HELP-SHAPE-THIS.md.
empathySync is not for people who want a companion AI or always-on assistant. It's for people who want useful help that doesn't try to become a habit.
The philosophy
We optimise for exit, not engagement.
| Practical Tasks | Sensitive Topics |
|---|---|
| Writing emails, coding, explanations | Emotional, health, financial, relationships |
| Full assistance, no limits | Brief responses, redirects to humans |
| Complete the task thoroughly | Encourage human connection |
- Crisis detection: immediate redirect to professional resources, no exceptions
- Post-crisis protection: never apologises for safety interventions
- Tracks dependency patterns and warns you if you're relying on it too much
- Transparency panel showing exactly why it responded the way it did
- Anti-engagement metrics: fewer sensitive sessions = success
- Suggests real humans to talk to, and helps you find them if you don't have anyone yet
git clone https://github.com/Olawoyin007/empathySync.git
cd empathySync
cp .env.example .env
docker compose upThis starts both empathySync and Ollama together. The model pulls automatically on first run. Open http://localhost:8501.
Any Ollama model works - qwen2.5:7b-instruct, llama3.1:8b, mistral:7b, whatever you already have. Set OLLAMA_MODEL in .env before running. Recommended: qwen2.5:7b-instruct for best quality.
git clone https://github.com/Olawoyin007/empathySync.git
cd empathySync
bash install.shThe install script checks Python, creates a virtual environment, installs dependencies, configures .env, and pulls the configured model automatically if Ollama is running. Then launch:
venv/bin/python -m streamlit run src/app.pygit clone https://github.com/Olawoyin007/empathySync.git
cd empathySync
pip install -e ".[dev]"
cp .env.example .env
empathysync # Launches Streamlit web UI
empathysync --mode cli # Direct terminal mode- Python 3.9+
- Ollama running locally (or via Docker)
- 8GB RAM recommended (4GB minimum with smaller models)
- GPU optional but improves response time
Lower-spec machine? Smaller models like qwen2.5:3b or tinyllama run comfortably on 4GB RAM. The safety pipeline remains intact regardless of model size.
- Local LLM: Runs entirely on your hardware via Ollama
- Privacy-First: Zero external API calls, complete data sovereignty
- Dual Interface: Streamlit web UI or direct CLI mode (
empathysync --mode cli) - Streaming Responses: Real-time token streaming for faster perceived response times
- YAML-Driven: All prompts, rules, and thresholds configurable
- LLM Classification: Optional intelligent classification for nuanced context detection
- Framework-Agnostic Core:
ConversationSessionclass can be embedded in any Python project - Transcript Export: Download any conversation as a Markdown file from the Session & Data panel
See .env.example for all configuration options:
# Required
OLLAMA_HOST=http://localhost:11434
OLLAMA_MODEL=qwen2.5:7b-instruct
OLLAMA_TEMPERATURE=0.7
# Optional
LLM_CLASSIFICATION_ENABLED=true # Intelligent context-aware classification
STORE_CONVERSATIONS=true # Local storage only
USE_SQLITE=false # SQLite backend (better concurrency)
ENABLE_DEVICE_LOCK=false # Multi-device sync safety- CLAUDE.md - Architecture and development guide
- ROADMAP.md - Detailed feature implementation plan
- MANIFESTO.md - Design principles and philosophy
- scenarios/README.md - Knowledge base editing guide
- docs/ - Additional documentation
empathySync is shaped by more than code. Engineers build the pipeline. But whether the words actually land - whether a response to "I feel lonely" feels human or hollow - that's a different kind of knowledge.
If you're a therapist, counsellor, social worker, or UX writer, see HELP-SHAPE-THIS.md. The responses, interventions, and connection-building guidance are plain text files. No programming required.
If you're an engineer, see CONTRIBUTING.md.
MIT License - Built for everyone's benefit and maximum accessibility.
The goal isn't a better chatbot. It's a world where you need chatbots less.