CH
Back to projects
Active
Governance DesignFull-Stack DevelopmentSystem ArchitectureDeveloper Experience

Agent Governance Console

Unified governance dashboard for multi-agent AI ecosystem

Full audit trailhash-chained tamper-evident logging across all agents
Dashboard Walkthrough

Full dashboard walkthrough: Overview, Audit Trail, Health, Governance, Intelligence, Planning, and Hooks pages

Product Screenshots
Agent Governance Console overview dashboard showing ecosystem health score, recent activity, and spoke status

Overview dashboard: ecosystem health, activity feed, and spoke status at a glance

AGC audit trail showing hash-chained entries with project filtering and impact classification

Tamper-evident audit trail with hash-chained logging and impact classification

AGC health monitoring dashboard showing per-factor breakdown and trend data

Health monitoring with per-factor breakdown across all spokes

Architecture
AGC DATA PIPELINEHOOKS (15)secrets-scannerrag-check · spawn-gate3 blocking · 12 advisoryMONITORINGalerts · escalationhealth · smoke testsINGESTIONaudit_entrieshealth_checkshook_invocationsgovernance_metricsdecisionsaccomplishmentsalerts · triage_logsintelligence_runsproposalsevery 15 minutesPOSTGRESQL28 tableshash-chained audit trailsession decision capturehook telemetryunified alertshash-chained integrityDASHBOARDOverviewAudit TrailAlertsHealthGovernanceIntelligencePipelinePlanningArchitectureNext.js + ElectronDATA SOURCES (ALL SPOKES)Retail Analyticsaudit · health · proposalsPortfolioaudit · health · proposalsB2B SaaSaudit · health · proposalsGaming Platformaudit · health · proposals
Problem

Multi-agent systems operating across multiple projects had no centralized visibility. Each spoke's audit trail, health status, and governance metrics were scattered. No way to know if agents were following protocols, if hooks were working, or if the ecosystem was healthy without manual investigation.

Discovery

After implementing governance hooks (secrets scanner, impact classification, audit logging), realized the data they generated had no unified presentation layer. Agents were producing hash-chained audit logs, L1-L4 impact classifications, and self-evaluation scores — but all stored in flat files across different repos. Tracked the gap between "generating governance data" and "making governance visible."

Solution

Built a Next.js dashboard backed by PostgreSQL (26 tables) that ingests data from all spokes. Key pages: Sessions (trace-level span visibility), Costs (token spend tracking), Intelligence (decisions, proposals, patterns), Einstein (multi-persona decision analysis), System (health, hooks, agents). Wrapped in an Electron menubar app for always-on access. Automated ingestion from all spokes via lifecycle hooks and nightly pipelines.

Decisions & Tradeoffs
  • PostgreSQL over SQLite — needed concurrent reads from multiple data sources, complex aggregations, and real-time ingestion from hooks
  • v2 rebuild after benchmarking against 9 production tools (LangSmith, Braintrust, Arize Phoenix) — stripped 17 over-engineered pages to 5 focused views backed by real data
  • Hash-chained audit trail — tamper-evident by design, each entry's hash includes the previous entry's hash
Outcomes
  • PostgreSQL backend ingesting governance data from project spokes in real-time
  • Hash-chained tamper-evident audit trail across all agent operations
  • Decision outcome tracking with confidence scoring and reasoning capture
  • Hook-based enforcement system ensuring protocol compliance across all agents
  • Nightly intelligence pipeline generating and scoring proposals with automated triage
  • Session trace view with span-level visibility into every tool call and agent spawn
  • Electron wrapper providing persistent menubar access with CSS-injected dark theme
Built with
Next.jsTypeScriptPostgreSQLElectronRadix UITailwind CSS