DecisionBox Documentation
Version: 0.1.0
DecisionBox is an open-source AI-powered data discovery platform. It connects to your data warehouse, runs autonomous AI agents that explore your data, and surfaces actionable insights and recommendations.
Who Is This For?
- Product managers who want data-driven insights without writing SQL
- Data analysts who want AI to augment their exploration
- Developers building data products who need automated pattern detection
- Game studios (our first domain pack) analyzing player behavior, churn, monetization
What Can It Do?
- Connect to BigQuery or Amazon Redshift (more warehouses coming)
- Run AI agents that write SQL, analyze results, and iterate autonomously
- Discover patterns across churn, engagement, monetization, and domain-specific areas
- Validate findings against actual data (not just LLM hallucination)
- Generate specific, numbered action steps — not generic advice
- Learn from your feedback — liked and disliked insights inform the next run
- Estimate costs before running (LLM tokens + warehouse query costs)
Documentation Structure
Getting Started
New to DecisionBox? Start here.
- Quick Start — Docker Compose to first discovery in 5 minutes
- Installation — All installation methods
- Your First Discovery — End-to-end walkthrough
Concepts
Understand how DecisionBox works.
- Architecture — System components and data flow
- Discovery Lifecycle — What happens during a discovery run
- Domain Packs — How domain-specific analysis works
- Providers — Plugin architecture for LLM, warehouse, and secrets
- Prompts — How AI prompts work, template variables, customization
Guides
Step-by-step instructions for common tasks.
- Creating Domain Packs — Build analysis for your industry
- Adding LLM Providers — Support a new LLM service
- Adding Warehouse Providers — Support a new data warehouse
- Adding Secret Providers — Support a new secret manager
- Configuring LLM Providers — Claude, OpenAI, Ollama, Vertex AI, Bedrock
- Configuring Warehouses — BigQuery, Redshift setup
- Configuring Secrets — Encrypted key management
- Customizing Prompts — Edit prompts, add analysis areas
- Project Profiles — How profiles improve insight quality
Reference
Detailed specifications.
- API Reference — All REST endpoints with examples
- Configuration — All environment variables
- CLI Reference — Agent command-line flags
- Prompt Variables — Template variable reference
- Data Models — Insight, Recommendation, Discovery models
- Makefile Targets — Build, test, dev commands
Deployment
Run DecisionBox in production.
- Docker Compose — Full deployment guide
- Production Considerations — Security, scaling, monitoring
Contributing
Help improve DecisionBox.
- Development Setup — Local dev environment
- Testing — Test suite and writing tests
- Pull Requests — PR process and conventions