dint 0.1.0

A personal AI tutor with long-term memory and tool use.

dint

dint is a Socratic AI tutor that teaches through guided discovery. Instead of giving you answers, dint asks the right questions to help you build understanding from the ground up.

Features

• Socratic dialogue — dint never just tells you the answer; it leads you there with questions
• Adaptive skill tracking — estimates your proficiency per concept and adjusts difficulty
• Knowledge graph — builds a map of concepts and their relationships as you learn
• Long-term memory — remembers your preferences, misconceptions, and learning goals across sessions
• Web search — looks up current information when needed
• Session management — multiple conversation threads, each with its own progress

Quick Start

Option A — pip install (recommended)

# 1. Install from the project root
pip install .

# 2. Configure your API key
cp .env.example .env
# Edit .env — add your OpenAI / Anthropic / other provider key

# 3. Run
dint

Option B — run.sh (dev mode with auto-reload)

chmod +x run.sh
./run.sh

Option C — uv

uv sync
uv run dint

The app will be available at http://localhost:7070.

CLI options

dint [--host HOST] [--port PORT] [--no-reload]

Flag	Description	Default
--host	Bind address	0.0.0.0
--port	Port to listen on	7070
--no-reload	Disable auto-reload (production)	(reload on)

Architecture

dint/
├── src/dint/
│   ├── app.py          # FastAPI application + REST routes
│   ├── agent.py        # Tool-calling orchestration loop
│   ├── llm.py          # LLM provider (OpenAI-compatible)
│   ├── tools.py        # Tool definitions & executors
│   ├── reflection.py   # Post-turn analysis (skills, memory, KG)
│   ├── persona.py      # System prompt / teaching persona
│   ├── db.py           # SQLite persistence layer
│   ├── config.py       # Environment configuration
│   ├── cli.py          # CLI entry point (argparse + uvicorn)
│   └── frontend/       # Bundled static assets
│       ├── index.html  # SPA shell
│       ├── style.css   # Dark-theme UI
│       └── app.js      # Client logic
├── frontend/           # Source frontend (dev copy)
├── run.sh              # One-command launcher (dev)
├── pyproject.toml      # Package metadata & dependencies
└── .env.example        # Config template

How It Works

1. You send a message via the web UI.
2. The agent assembles context: persona, relevant memories, skill estimates, and knowledge subgraph.
3. The LLM responds, optionally calling tools (web search, memory recall, skill lookup, KG query).
4. After the reply, the reflection engine analyses the exchange:
   • Updates skill confidence scores
   • Extracts new knowledge-graph nodes/edges
   • Stores durable memories (preferences, corrections, goals)
5. The UI renders the reply and updates the side panels.

Configuration

Variable	Description	Default
OPENAI_API_KEY	API key for your provider	(required)
OPENAI_BASE_URL	OpenAI-compatible endpoint	https://api.openai.com/v1
DINT_MODEL	Model name	gpt-4o-mini
DATABASE_URL	SQLite database path	dint.db
SERPER_API_KEY	Serper.dev key for web search	(optional)

Requirements

• Python 3.11+
• An OpenAI-compatible API key (OpenAI, Anthropic via proxy, Ollama, etc.)
• (Optional) A Serper.dev key for web search

License

MIT