The Open-Source Multi-Agent Observability & Governance Engine
Trace hierarchical agent swarms, trap infinite tool loops with automatic cost circuit breakers, block prompt injections, and query real-time telemetry via native Model Context Protocol (MCP).
import agentwatch
from agentwatch import trace_agent, trace_llm, trace_tool
# Auto-instruments multi-agent chains & propagates parent context
with trace_agent("support_orchestrator", end_user_id="cust_9921"):
with trace_tool("query_knowledge_base") as tool:
kb_results = search_docs("How to configure SAML SSO?")
tool.finish(output={"matched_docs": 3})
with trace_llm("gpt-4.1-mini") as llm:
response = call_llm(kb_results)
llm.finish(output=response){
"input": "Customer asks for automated refund policy under compliance consent #cs_901.",
"output": "Refund processed under verified DPDP consent. Verification token: 0x88f9."
}Built for Autonomous Multi-Agent Swarms
Everything you need to observe, govern, and secure AI agents in production without burning unnecessary cloud spend.
Interactive Agent Topology Graph
Automatically maps delegation hierarchies, call volume, edge latencies, and error clusters across orchestrators, planners, and sub-agents.
Cost Circuit Breaker
Hard killswitch that stops runaway agents stuck in infinite loops ($50 in 5 min) and dispatches emergency PagerDuty webhooks.
● Status: ARMED & ACTIVE
Native MCP Server
Connect Claude Desktop, Cursor, or Windsurf directly via stdio JSON-RPC. AIs can query traces, check prompt security, and run evaluations.
$ agentwatch mcp --port 8000SHA-256 Audit Chains
Cryptographically verify span immutability and compliance actions with tamper-evident SHA-256 hash chains.
Hash: a4f89d02...
Prompt Injection Shield
Autonomous heuristic & semantic detectors trap jailbreaks, delimiter injections, and role overrides before reaching the model.
How NetrAI Compares
Engineered specifically for autonomous multi-agent swarms with enterprise safety and zero vendor lock-in.
| Capabilities | NetrAI | Langfuse | Maxim AI | LangSmith |
|---|---|---|---|---|
| Multi-Agent Topology GraphDirected agent network graph with call volume & error clusters | ||||
| Automated Cost Runaway Circuit BreakerHard killswitch for infinite tool loops & spend spikes | ||||
| Native Model Context Protocol (MCP) ServerDirect integration for Claude Desktop, Cursor & Windsurf | ||||
| Tamper-Evident SHA-256 Audit Log ChainsCryptographic hash chaining & append-only DB hardening | ||||
| Heuristic & Semantic Prompt Injection DefensePre-execution pattern matching & semantic LLM judge classifier | ||||
| Zero-Cost Local Mode (No Paid Keys Needed)Full tracing with local Ollama, Groq, or mock replay | PARTIAL | |||
| DPDP / GDPR Subject Rights Erasure WorkflowTwo-step admin confirmation gate & ClickHouse purge | ||||
| CI/CD Golden Dataset Regression GuardPre-deploy CLI evaluation with exact, semantic & judge tests |
Estimate Swarm Spend & Runaway Protection
Calculate your monthly LLM token volume and simulate how NetrAI prevents catastrophic runaway loop bills.
10.4M spans/mo
12.44B tokens
Circuit breaker triggers within 50ms if loop frequency surpasses 30 tool calls or spend exceeds $50.00.
Ready to Observe & Secure Your Agent Swarms?
Start tracing in less than 3 minutes. Zero credit card required. Run 100% free locally with Ollama or connect your production swarms.