Agent Discovery & Inventory
Find and catalog every AI agent in your environment — whether it uses the NodeLoom SDK or not.
Business & Enterprise
What It Does
The Agent Inventory gives you a single view of every AI agent across your organization. It tracks which models each agent uses, which tools it calls, and how risky it is — so you can enforce governance on agents you didn't even know existed.
Discovery Sources
Agents are discovered through five channels:
| Source | How It Works | What You Get |
|---|---|---|
| SDK Telemetry | Agents instrumented with the NodeLoom SDK are auto-cataloged as they send traces | Full observability: traces, models, tools, tokens, anomalies, guardrails |
| Cloud Providers | Connect AWS/Azure/GCP credentials to enumerate Bedrock agents, OpenAI deployments, Vertex endpoints | Agent names, deployed models, cloud metadata |
| GitHub Scanning | Scan your repos for AI framework imports (LangChain, CrewAI, OpenAI SDK, etc.) | Repo name, detected framework, language |
| MCP Gateway | Route MCP traffic through a NodeLoom proxy to observe tool calls without any SDK | Tool names, call frequency, success rates, latency |
| eBPF Kernel Monitoring | Deploy zero-instrumentation probes that intercept LLM API calls at the kernel level via SSL_write/SSL_read uprobes | Process names, LLM endpoints, token usage, timing — all without modifying any application code |
Agent Lifecycle
Every discovered agent goes through a status lifecycle:
| Status | Meaning |
|---|---|
| DISCOVERED | Just found — nobody has reviewed it yet |
| VERIFIED | A human reviewed and confirmed it's a legitimate agent |
| MONITORED | Linked to a workflow with active monitoring, guardrails, and compliance |
| ARCHIVED | No longer in use |
Risk Scoring
Each agent gets an auto-calculated risk score from 0 to 100 based on its governance posture:
- Guardrails configured — reduces risk by 20 points
- Monitoring enabled — reduces risk by 15 points
- Red team scan passed — reduces risk by 15 points
- Data source dependencies — increases risk by 10 points each
- Recent anomalies — increases risk by 5 points each
- Guardrail violations — increases risk by 10 points each
Scores: Low (0-30), Medium (31-60), High (61-100).
Dependency Graph
The dependency graph visualizes relationships between agents, models, tools, and MCP servers. Navigate to Agent Inventory → Graph to see which agents share models, which tools are used most, and where your risk is concentrated.
Onboarding Discovered Agents
When you discover an agent that isn't using the SDK yet, click the Onboard button on its detail page. You'll get language-specific setup instructions with your SDK token pre-filled. Once the agent starts sending telemetry, it automatically transitions to MONITORED status with full observability.