An AI SRE copilot for the Agents of SigNoz

Your AI agent's failures, explained.

Cerberus reads your agent's SigNoz traces, ranks every run by failure and spend, explains the worst ones with trace citations, and writes the alert that would have caught them.

Three heads, one watchdog

01

Observe

Every agent run lands in SigNoz as OpenTelemetry traces — token usage, cost, latency, and errors, per step. Cerberus emits the telemetry and reads it back.

02

Explain

The copilot reads the ranked incidents and answers in plain English, grounded only in the facts it was given, and cites the trace_id. No speculation.

03

Prevent

One click writes the SigNoz alert rule that would have caught the incident — created through the same MCP server Cerberus reads from.

How it works

Cerberus both emits telemetry to SigNoz and consumes it back — the read path speaks MCP, so the same tool surface an AI client gets is the surface Cerberus is built on.

Agent
OpenTelemetry
SigNoz
MCP server
Cerberus
Alert rule

Reproducible in one command

The repo ships casting.yaml and its lock. One Foundry command brings up SigNoz and the MCP server — no click-built setup to reproduce.

# bring up SigNoz + the MCP server

$ foundryctl cast -f casting.yaml

# read incidents, explain, arm the alert

$ uvicorn cerberus.api:app

SigNoz · OpenTelemetry · SigNoz MCP server · FastAPI · Next.js