ahy-governance 0.8.0

Multi-Agent Governance Platform — Conflict Detection, Cost Tracking, Audit Logging

Ahy Governance

Open-source AI Agent governance. Deploy agents with confidence — conflict detection, cost control, audit trails, compliance ready.

中文文档

Recent Updates

• Web Dashboard — one-command launch, dark monitoring theme, 7 live-updating panels
• Transparent Proxy — OpenAI-compatible endpoint, zero code changes for existing agents
• Built-in Auth — email/password login + JWT + API Key management
• Docker Deploy — single command, supports Zeabur / EdgeOne Pages

What You Get

Problem	Ahy Governance
Agents conflicting with each other?	Auto-detect and resolve — 5 conflict types
AI costs unpredictable?	Real-time budget visibility + automatic circuit breaker
Auditor knocking?	5-minute compliance evidence export (SOC2 / ISO27001)
Someone jailbreaking your AI?	Prompt injection detection + PII redaction
Intern fat-fingering production?	3-tier RBAC — who can view, who can change

---

Before vs After

	Without ahy-governance	With ahy-governance
Ops burden	1 person, 2 hrs/day monitoring	10 min/day reading summaries
Conflict incidents	3-5 business errors per month	Auto-detect + resolve, 0 incidents
Compliance audit	2 weeks manual log gathering	One-click PDF export, 10 minutes
AI spending	Unpredictable, over budget monthly	Real-time visibility + auto circuit breaker
Access control	Permission chaos	Self-service role management

---

3 Lines to Integrate

from ahy_governance import ConflictDetector, CostTracker, AuditReporter

detector = ConflictDetector(); tracker = CostTracker(); auditor = AuditReporter()

---

Beyond LLM Observability

LangSmith and LangFuse are excellent LLM tracing tools, but they don't understand multi-agent orchestration. Ahy Governance is purpose-built for teams running 5+ agents that collaborate, conflict, and need coordination.

Capability	Ahy Governance	LangSmith	LangFuse	Datadog
Multi-agent conflict detection	✅ 5 types	❌	❌	❌
Cross-agent cost attribution	✅ Per-agent	Partial	Partial	❌
Tamper-proof audit (SHA-256)	✅ SOC2/ISO	❌	❌	Partial
RBAC + API key management	✅ 3-tier	❌	❌	✅
Prompt injection defense	✅ 13 rules	❌	❌	❌
Cross-agent memory sharing	✅ Namespaced	❌	❌	❌
Open source	✅ MIT	❌	✅ MIT	❌

---

Who's It For

• You're building agents with LangChain / CrewAI and your boss asks "is this safe?"
• Compliance audit is coming and auditors want every AI decision traceable
• You manage 20+ agents and don't know who spent what or who's conflicting with whom
• You need SOC2 / ISO27001 evidence and Excel isn't cutting it anymore

---

Quick Start

pip install ahy-governance[web]
ahy-dashboard
# Open http://localhost:8080 — click "Demo Data" to populate

Or use individual modules:

from ahy_governance import ConflictDetector, CostTracker, AuditReporter

# Detect conflicts between agents
detector = ConflictDetector()
conflicts = detector.check(agent_outputs, dag_definition)

# Track costs per agent
tracker = CostTracker()
tracker.set_budget(limit_usd=100)
tracker.track("Planner", "claude-opus-4-7", tokens_in=15000, tokens_out=8000)

# Tamper-proof audit logging
auditor = AuditReporter()
auditor.log(AuditEventType.AGENT_START, "Planner", {"task": "plan"})

---

Web Dashboard

Launch with one command. 7 panels, dark theme, auto-refresh.

ahy-dashboard

Panel	What it shows
Dashboard	Agent health overview, total cost, audit integrity, budget gauge
Health	Per-agent status badges, P50/P95/P99 latency, success rates
Cost	Budget gauge, cost by agent/model, per-call entry log
Conflicts	JSON sandbox — paste outputs + DAG, click "Check"
Audit	Hash-chained event log, integrity verification, SOC2/ISO27001 export
Memory	Namespace browser, key-value search, cross-agent shared state
Security	RBAC workspace/user/key management + Prompt Guard sandbox

---

Architecture

┌──────────────────────────────────────────────────────┐
│              Ahy Governance Dashboard                 │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐  │
│  │ Conflict │ │   Cost   │ │  Audit   │ │ Health │  │
│  │ Detector │ │ Tracker  │ │ Reporter │ │Monitor │  │
│  └──────────┘ └──────────┘ └──────────┘ └────────┘  │
├──────────────────────────────────────────────────────┤
│                 Governance Core                       │
│  ┌──────────┐ ┌──────────┐ ┌──────────────────────┐  │
│  │  Memory  │ │   RBAC   │ │    Prompt Guard      │  │
│  │ Sharing  │ │          │ │ (Injection + PII)    │  │
│  └──────────┘ └──────────┘ └──────────────────────┘  │
├──────────────────────────────────────────────────────┤
│      Existing Agent Core (not included — bring your own)  │
│      Orchestrator  │  TraceLogger  │  Router         │
└──────────────────────────────────────────────────────┘

---

Modules (7/7 complete)

#	Module	Tests	Description
1	Conflict Detector	23	5 conflict types: fact, format, dependency, scope, confidence
2	Cost Tracker	46	22 model pricings, budget circuit breaker, per-agent attribution
3	Audit Reporter	35	SHA-256 hash chain, SOC2/ISO27001 compliance export
4	Health Monitor	45	Heartbeats, P50-P99 latency, error rates, DAG pipeline tracking
5	Auth & RBAC	59	Email/password + JWT, 3-tier roles, API key lifecycle
6	Prompt Guard	39	13 injection patterns, PII redaction, sanitize pipeline
7	Memory Sharing	34	Namespaced key-value, TTL expiry, tag search

384 tests, 0 failures.

---

SOC 2 / ISO 27001 Compliance

The Audit Reporter module provides cryptographic audit logging:

• SHA-256 hash chain — every audit entry is cryptographically linked to its predecessor
• SOC 2 export — one-click report covering Security, Availability, Confidentiality, Processing Integrity, and Privacy
• ISO 27001 export — Annex A controls with compliant/needs-review status per control

---

Contact

Questions or feedback? Open an issue or email [e-mail].

---

Ecosystem

Project	Description	Status
Kingdee MCP Server	AI Agent ↔ 金蝶云星空 ERP	✅ MIT
WeCom MCP Server	AI Agent ↔ 企业微信	✅ MIT
Ahy Agent	Multi-agent orchestration harness	v0.6.0

---

Community

• Discussions: GitHub Discussions — questions, ideas, feedback
• Issues: GitHub Issues — bug reports, feature requests
• Star the repo: If this is useful, a star helps others discover it

---

Contributing

PRs welcome. See CONTRIBUTING.md.

---

MIT License. Built by Leo-Ayh-Oday.