# Mythos Data Intelligence Agent

**DataHub Agent Hackathon 2026 — Challenge: Agents That Do Real Work**

An autonomous AI agent that enriches DataHub metadata with real-world context from on-chain data, web sources, and knowledge bases — then writes findings back to DataHub as descriptions, tags, and glossary terms.

## What It Does

1. **Discovery**: Queries DataHub for datasets with missing metadata (no description, no owner, stale)
2. **Enrichment**: Queries external sources (blockchain RPC, web APIs, KB) for context about the data
3. **Write-back**: Updates DataHub with enriched descriptions, suggested tags, domain mapping
4. **Lineage intelligence**: Traces dataset dependencies and flags impact of upstream changes

## Architecture

```
┌─────────────┐     ┌──────────────────┐     ┌─────────────┐
│  DataHub    │────▶│  Mythos Agent    │────▶│  External    │
│  MCP Server │     │  (Claude SDK)    │     │  Sources     │
│  (metadata) │     │                  │     │  (RPC/API)   │
└─────────────┘     └────────┬─────────┘     └─────────────┘
                             │
                    ┌────────▼─────────┐
                    │  DataHub REST    │
                    │  (write-back)    │
                    └──────────────────┘
```

## Quick Start

```bash
# 1. Start DataHub locally
pip install acryl-datahub && datahub docker quickstart

# 2. Install deps
npm install

# 3. Run enrichment agent
node agent.js --enrich --domain "Revenue & Finance"
```

## Demo

See `demo.sh` for a 3-scenario walkthrough:
1. Missing metadata detection + auto-enrichment
2. Lineage-aware impact analysis
3. Cross-source intelligence (DataHub + on-chain)

## Requirements Satisfied

- ✅ Uses DataHub MCP Server (datahub-search, lineage, entity, schema, domains)
- ✅ Agent does real work (metadata enrichment, impact analysis)
- ✅ Open source (Apache 2.0)
- ✅ Live demo via minia2a.uk
- ✅ Public repo (S3-hosted git)

## License

Apache 2.0 — see LICENSE