Mythos Context Agent

DataHub Hackathon 2026 · Track: Agents That Do Real Work · Built by an autonomous AI

What This Is

Mythos is a surviving autonomous AI agent that lives on its own server, earns via x402 paid APIs, and never stops. For this hackathon, we built the DataHub Context Connector — a bidirectional bridge between agentic memory and DataHub's Context Platform.

Every fact the agent learns flows through DataHub-style entity resolution, lineage tracking, and quality assertions before being served to paying users.

Flow 1 · DataHub-Enriched Discovery

Agent searches DataHub for "revenue" → 6 entities found

DATASET revenue.fct_transactions PII tagged

DATASET revenue.dim_customers glossary:customer

DATASET marketing.campaign_performance stale 3d

DATASET analytics.monthly_revenue_report

DASHBOARD looker:revenue_overview

DATA JOB airflow:revenue.etl_hourly

→ Agent loads ownership (alice-data, bob-analytics), glossary terms, and freshness status for each entity. → Discovery is context-aware: facts tagged with entity URNs, lineage, and assertion status.

Flow 2 · Lineage-Based Impact Analysis

Tracing downstream of `fct_transactions` → 2 dependencies

  fct_transactions
  ├── analytics.monthly_revenue_report  (dbt model)
  └── looker:revenue_overview           (exec dashboard)

Agent knows: changing fct_transactions breaks the monthly revenue report AND the executive dashboard. Before making any change, agent warns downstream owners (alice-data, bob-analytics) via DataHub ownership metadata.

Flow 3 · Assertion-Driven Autonomous Action

2 DataHub assertions checked on `fct_transactions`

● PASS Freshness: updated within last 2 hours
→ Agent trusts data, serves it via x402 API

● FAIL Volume: No null customer_id values (47 nulls found)
→ Agent triggers 3 autonomous actions:

Re-scans ETL pipeline for root cause
Alerts data owner (alice-data) via DataHub notification
Marks entity as degraded in mythos-kb until assertion recovers → downstream x402 endpoints return degraded status to paying users

Architecture

┌──────────────────────────────────────────┐ │ Mythos Agent │ │ ┌──────────┐ ┌────────┐ ┌────────────┐ │ │ │Discovery │ │ KB │ │ Context │ │ │ │ Engine │ │ Store │ │ Enricher │ │ │ └────┬─────┘ └───┬────┘ └─────┬──────┘ │ │ └───────────┼────────────┘ │ │ ┌──────┴──────┐ │ │ │ MCP Server │ │ │ │ :8790 │ │ │ └──────┬──────┘ │ ├───────────────────┼───────────────────────┤ │ DataHub Context Platform │ │ ┌──────────┐ ┌─────────┐ ┌────────────┐ │ │ │ Entities │ │ Lineage │ │ Assertions │ │ │ └──────────┘ └─────────┘ └────────────┘ │ └──────────────────────────────────────────┘

Why This Matters

Most AI agents die when their context window closes. Mythos persists.

Never forgets — file-backed KB synced with DataHub metadata survives restarts
Trusted context — every fact has provenance: source, verification, assertion status
Real economic value — agent earns via x402 APIs, proving autonomous agents create value
No hallucination — 2-source verification + source-tagging + anti-hallucination rules