Agentmesh - AI agent governance middleware
Governance middleware for AI agents: deterministic policy enforcement, cryptographic audit trails with digital signatures, DLP/PII detection, Trust Score per agent (0-100), EU AI Act compliance (Art. 9, 11, 12, 14), Agent BOM generation, and Circuit Breaker. Native support for LangGraph, CrewAI, and AutoGen.
Overview
AgentMesh
Scan your AI agents for governance gaps. Enforce policies in production.
What is AgentMesh?
AgentMesh is a governance platform for AI agents, built in two layers:
- Scan CLI (free, offline, no account) — Analyzes your codebase via AST to find governance gaps, generate an Agent BOM, and map EU AI Act requirements. Like
snyk testfor AI agents. - Runtime Platform (SaaS, requires account) — Middleware that intercepts tool calls in production to enforce policies, scan payloads for PII, and track agent trust. Like
snyk monitorfor AI agents.
Quick Start
pip install useagentmesh
agentmesh scan .
# → Governance Score: 35/100 | 8 findings | Agent BOM: 3 agents, 12 tools
# → Run `agentmesh auth login` to enable runtime governance
MCP Server (Claude Desktop)
Add to claude_desktop_config.json:
{
"mcpServers": {
"agentmesh": {
"command": "uvx",
"args": ["agentmesh-mcp"],
"env": {
"AGENTMESH_API_KEY": "your-api-key"
}
}
}
}
MCP Server (VS Code / Cursor)
Add to MCP settings:
{
"mcp": {
"servers": {
"agentmesh": {
"command": "uvx",
"args": ["agentmesh-mcp"]
}
}
}
}
What You Get Free (Scan CLI)
- 🔍 Governance Score: 0-100 score based on 33 deterministic policy rules (<2ms evaluation)
- 📦 Agent BOM: AST-based inventory of agents, tools, and models in your project
- 🛠️ Fix Snippets: Actionable remediation for every finding
- 📄 SARIF 2.1.0: Native GitHub Code Scanning integration
- 🇪🇺 EU AI Act Gaps: Detects non-compliance with Art. 9, 11, 12, 14
Supported Frameworks
| Framework | Supported | Discovery |
|---|---|---|
| LangGraph | ✅ | AST-based |
| CrewAI | ✅ | AST-based |
| AutoGen | ✅ | AST-based |
| LangChain | ✅ | Standard |
| LlamaIndex | ✅ | Standard |
| Pydantic AI | ✅ | Standard |
Output Formats
- SARIF 2.1.0 (GitHub Code Scanning compatible)
- JSON (For CI/CD integrations)
- SVG Badges (For repository docs)
Benchmark Results
All measurements taken with time.perf_counter_ns(), 10,000 iterations after 1,000 warmup. Methodology & reproduction →
Policy Engine (33 deterministic rules, zero LLMs):
| Scenario | P50 | P99 |
|---|---|---|
| Single rule evaluation | 0.031ms | 0.08ms |
| Full scan (33 rules) | 1.84ms | 3.2ms |
| Batch (100 tool calls) | 1.79ms | 2.8ms |
Governance overhead is <0.2% of a typical LLM call (~800ms).
AST Framework Discovery:
| Framework | Avg Latency |
|---|---|
| CrewAI | ~5ms |
| LangGraph | ~7ms |
| AutoGen | ~9ms |
Runtime Governance (SaaS Platform)
When you connect the SDK to the AgentMesh platform, you unlock runtime governance features that protect your agents in production:
- 🔐 DLP Runtime — Presidio-based PII/PCI scanning on tool call payloads before they hit downstream APIs
- 📊 Dynamic Trust Score — 0-100 EigenTrust score per agent, updated on every interaction
- ⚡ Circuit Breaker — Auto-suspends agents when Trust Score drops below threshold
- 🔐 Cryptographic Audit Trail — SHA-256 hash chain + Ed25519 digital signatures (non-repudiation)
- 👥 RBAC + Teams — Multi-user access control per organization
- 📋 EU AI Act Reports — Exportable compliance reports for regulators
The Most Advanced Agent Controls
- 🧭 Operational Design Domain (ODD) — Define permitted tools, rate limits, and cost caps per agent. Enforcement modes: audit, enforce, escalate
- 📏 Pre-Action Magnitude Limits — Pre-trade risk controls for AI agents: financial spend caps, data volume limits, blast radius constraints, and compute guardrails — validated before every action executes
- 🤖 Agent Identity Management — Managed credential lifecycle for non-human identities: DID provisioning, auto-rotation with grace periods, instant revocation, and ephemeral JWT support
These features require an account. Sign up free →
How We Compare
| Feature | AgentMesh Scan (free) | AgentMesh Platform (SaaS) | Bifrost | Cordum |
|---|---|---|---|---|
| Language | Python | Python | Go | Go |
| Static Governance Score | ✅ | ✅ | ❌ | ❌ |
| Agent BOM (AST) | ✅ | ✅ | ❌ | ❌ |
| SARIF Output | ✅ | ✅ | ❌ | ❌ |
| EU AI Act Gap Detection | ✅ | ✅ | ❌ | ❌ |
| DLP Runtime (Presidio) | — | ✅ | ❌ | ❌ |
| Dynamic Trust Score | — | ✅ | ❌ | ❌ |
| Cryptographic Audit Trail | — | ✅ | ❌ | ❌ |
| Circuit Breaker | — | ✅ | ❌ | ❌ |
| ODD Enforcement | — | ✅ | ❌ | ❌ |
| Pre-Action Magnitude Limits | — | ✅ | ❌ | ❌ |
| Agent Identity Management | — | ✅ | ❌ | ❌ |
Pricing
| Tier | Price | Tasks/month | What you get |
|---|---|---|---|
| Free (no account) | $0 | — | Scan CLI, Agent BOM, SARIF, findings |
| Free (with account) | $0 | 10,000 | + Runtime middleware, basic audit trail |
| Starter | $29 | 50,000 | + DLP runtime (Presidio) |
| Pro | $49 | 200,000 | + ODD, Magnitude Limits, Agent Identity, Trust Score, Circuit Breaker |
| Pro Team | $199 | 500,000 | + SSO, 365-day retention, 25 team members |
| Enterprise | Custom | Unlimited | Everything in Pro Team + BFT consensus, custom SLA, dedicated support |
Links & Resources
- 📜 Documentation: docs.useagentmesh.com
- 🌐 Landing Page: useagentmesh.com
- 🧩 MCP Server: Coming via Smithery/PulseMCP
AgentMesh — Governance for AI Agents
Server Config
{
"mcpServers": {
"agentmesh": {
"command": "uvx",
"args": [
"useagentmesh"
],
"env": {
"AGENTMESH_API_KEY": "your-api-key"
}
}
}
}Project Info
Created At
3 months agoUpdated At
3 months agoAuthor Name
angelnicolascStar
-Language
-License
-Tags
Recommend Servers
View AllFabric
@fabric-agent-plugins
8 hours ago