Tell it what happened. It maps your rights, generates the question, and gives you the next step.

Local-first. No data leaves your device. Built on the Burgess Principle (UK00004343685).

MIT licence. Python ≥ 3.11.

• Classifies your situation from a plain-English description into one of nine domains: enforcement, benefits, housing, platform, medical, credit, employment, immigration, and consumer.
• Maps your rights — shows the specific legal rights that apply, with sources and actions.
• Gives you one next step — not a list, just one clear thing to do now.
• Drafts the communication — 27 ready-to-send letter templates (SAR requests, FOI requests, grievance letters, NHS complaints, mandatory reconsiderations, tribunal appeals, ICO complaints, chargeback requests, and more), each with the Burgess Principle question included.
• Tracks deadlines — statutory timelines for DSARs, FOIs, tribunal claims, and complaint acknowledgements.
• Hashes every outgoing message — SHA-256 commitment hashing so you can prove a message came from you, without revealing personal data.
• Optional local AI — connect a local AI backend (Ollama) for adaptive classification, context-aware rights mapping, and situationally tailored Burgess questions. Disabled by default. When enabled, all AI processing stays on your device.

Everything runs on your device. No server, no accounts, no tracking, no cookies.

pip install -e ".[dev]"
pytest

Open web/index.html in your browser. No server needed.

/core        Python modules — conversation, rights, next step, templates, commitment, timeline, AI adapter
/prompts     System prompts for local AI integration (master-system.md)
/templates   Ready-to-send letter templates (27 templates)
/web         Local-first chat interface (HTML + CSS, no dependencies)
/docs        User guide, security policy, contributing guide

• User guide — full usage instructions, configuration, and examples
• Contributing — how to help, code style, and pull request process
• Security policy — vulnerability reporting, data handling, and threat model

Mirror is local-first by design.

• No data leaves your device
• No accounts or registration
• No analytics or tracking
• No cookies
• The web interface makes no network requests
• Your commitment vault is encrypted on your device
• AI features (when enabled) use a local backend — no cloud APIs

Mirror can optionally connect to a local AI backend for adaptive classification, context-aware rights mapping, and situationally tailored Burgess questions. AI is disabled by default — the deterministic rule-based engine works without it.

1. Install Ollama and pull a model:

   ollama pull mistral

2. Enable AI in mirror-config.json:

   {
     "ai": {
       "enabled": true,
       "model": "mistral",
       "base_url": "http://localhost:11434",
       "timeout": 120
     }
   }

3. Install the optional AI dependency:

   pip install -e ".[ai]"

from core.ai_adapter import adaptive_classify, adaptive_burgess_question, is_backend_available

# Check if the backend is running
if is_backend_available():
    # AI-powered classification
    result = adaptive_classify("My PIP assessment was refused")

    # Situational Burgess question
    question = adaptive_burgess_question("My PIP was refused", "benefits")

When AI is unavailable or disabled, these functions return None, and Mirror falls back to its deterministic engine. Nothing breaks.