Open Source

Manage AI Agents
Like HR Manages People

The first open-source Agent System of Record. Onboard, organize, monitor, govern, and pay your AI agent workforce—using the same HR/ERP concepts you already know.

Explore Features View Architecture
AgentHR - AI Agent Network Visualization

Built on open standards

MCP Protocol
A2A Protocol
PostgreSQL + TimescaleDB
Qdrant Vector DB
Ollama Local AI
Features

Everything You Need to Run an Agent Workforce

Six integrated domains covering the full lifecycle of AI agent management—from hiring to retirement.

Agent Directory & Profiles

6-section agent profiles with versioned system prompts, tool assignments, knowledge bases, constraints, and relationship mapping. Full lifecycle from draft to retirement.

Task Orchestration

Hierarchical goal trees, tiered task classification, agent-to-task assignment with atomic checkout, heartbeat scheduling, and conductor workflow visualization.

Cost & Budget Control

Per-agent, per-team, and org-wide budgets with real-time tracking. Model cascade (Opus to Haiku) on budget pressure. Compensation plans and spend forecasting.

Governance & Trust

5-tier trust levels (Intern to Executive), constitutional contracts with monotonically decreasing privileges, 3-tier tool classification, and policy enforcement under 200ms.

Metrics & Performance

TimescaleDB-powered time-series metrics, continuous aggregates, 1000-point code quality scores, performance reviews, and real-time dashboards via SSE.

Persistent Memory

Vector memory with Qdrant, 15 automated maintenance jobs, knowledge graph visualization, semantic search, dedup, consolidation, and context guard protection.

Workforce Management

Your Agent Directory, Organized

Every agent gets a structured profile—just like an employee record. Track capabilities, assign roles, manage teams, and visualize your entire agent organization at a glance.

  • Auto-discover agents from Claude Code, Codex, Gemini, and Ollama
  • Onboarding wizard: draft, configure, test, deploy
  • Version control with diff and one-click rollback
  • Visual org chart with drag-to-reorganize
Agent Directory Dashboard
Governance & Compliance

Trust, but Verify. Automatically.

Constitutional governance with immutable audit trails. Every tool call checked in under 200ms. Every delegation verified through 6 sequential gates. Full compliance mapping to ISO 42001, EU AI Act, and OWASP.

  • SHA-256 chained audit log—tamper-proof, 7-year retention
  • Human-in-the-loop approvals with SLA tracking
  • Monotonically decreasing privilege chains
  • 10 verification gates from CISO review to pre-release
Governance and Trust Visualization
Intelligence

Memory That Learns and Grows

Persistent vector memory with automated maintenance. Your agents build institutional knowledge over time—complete with a force-directed knowledge graph you can actually explore.

  • Semantic search across all memory collections
  • 15 automated maintenance jobs (dedup, consolidation, pruning)
  • RAG injection at agent spawn—agents start with context
  • 4-tier context guard prevents context window emergencies
Knowledge Graph Visualization
The HR Metaphor

Concepts You Already Know

Every AgentHR concept maps directly to familiar HR/ERP terminology. No new mental models required.

HR Concept
In AgentHR
Employee Profile
6-section Agent Profile (roles, tools, knowledge, constraints)
Onboarding
Step-by-step wizard: draft, configure, test, deploy
Org Chart
Visual agent hierarchy with departments and teams
Job Description
Versioned system prompt template with approval workflow
Performance Review
Quality scores, task completion, cost efficiency, error rates
Payroll
Token cost tracking with budgets and model cascade
Work Schedule
Heartbeat cron with execution windows and timezone
Trust Level
Intern (L1) to Executive (L5) permission tiers
Headcount
Active + paused + testing agent count per org
Retirement
Soft-delete lifecycle: deprecated, then retired
~55
Database Tables
266
Structured Requirements
8
Domain Models
3
External Dependencies
Technology

Modern Stack, Minimal Dependencies

Three external services. Everything else built-in. No n8n, no Supabase, no Redis.

Hono
API Server
React + Vite
Frontend
Drizzle ORM
Database Layer
PostgreSQL
Relational + JSONB
TimescaleDB
Time-Series Metrics
Qdrant
Vector Memory
Ollama
Local Embeddings
Better Auth
Multi-Tenant Auth
TanStack
Router + Query
shadcn/ui
UI Components
ReactFlow
Workflow DAG
Recharts + D3
Charts & Graphs
Architecture

Monolith With Process Isolation

Single process serves everything. Agent CLI subprocesses run in isolated process groups with resource limits and kill switches.

agenthr/
├── packages/
│  ├── server/ ← Hono API + SSE + scheduler + agent manager
│  ├── ui/ ← Vite React (static build served by server)
│  ├── db/ ← Drizzle schema + migrations + TimescaleDB
│  ├── memory/ ← Qdrant client + memory tools + maintenance
│  ├── governance/ ← Trust broker + policy engine + audit bus
│  ├── adapters/ ← Claude / Codex / Gemini / Ollama
│  ├── scheduler/ ← Cron + heartbeat engine
│  ├── vault/ ← AES-256-GCM credential store
│  ├── quality/ ← Code assurance orchestrator (67 tools)
│  ├── discovery/ ← System scanner (plugins, hooks, MCP, CLIs)
│  └── shared/ ← Types, utils, constants
├── docker-compose.yml ← PostgreSQL + TimescaleDB, Qdrant
└── pnpm-workspace.yaml

Ready to Manage Your Agent Workforce?

Open-source. Self-hosted. Three dependencies. Your Agent System of Record, running in minutes.

Coming Soon