opensre

Overview

opensre is Tracer-Cloud’s open-source framework for building AI Site Reliability Engineering agents. The core use case: when something breaks in production, the evidence is scattered across logs, metrics, traces, runbooks, and Slack threads. A human SRE has to correlate all of that, often at 2am. opensre automates the investigation phase — connecting to your existing observability stack, correlating signals, and surfacing a root-cause hypothesis with evidence links before you have to touch anything.

Key Capabilities

Correlated root-cause analysis: opensre ingests alerts, then automatically queries the relevant logs, metrics, and traces across your observability tooling. Rather than presenting raw data, it synthesizes a structured root-cause hypothesis with supporting evidence, similar to how a senior SRE would work through an incident.

Runbook-aware reasoning: The agent applies your existing runbooks during incident investigation — not as static documentation lookup, but as structured reasoning that adapts to the specific failure signals present. If the runbook says “check Postgres connection pool when error rate exceeds 5%,” the agent does exactly that in context.

Predictive detection: Beyond reactive incident response, opensre includes predictive failure detection — identifying patterns in metrics and logs that precede known failure modes before they trigger pages.

Synthetic test suites: opensre ships with end-to-end testing environments that simulate realistic infrastructure failure scenarios, letting you validate agent behavior before putting it on actual production on-call rotation.

Architecture

Python (99%), Apache 2.0, self-hosted on your own infrastructure. Connects to your existing Kubernetes clusters, cloud accounts, and observability platforms — it does not require routing production traffic through Tracer’s services. The LLM backend is pluggable: Anthropic, OpenAI, Gemini, and local Ollama are all supported, which matters for teams with data residency requirements.

Who It’s For

Platform and SRE teams handling production incidents who want to automate the initial triage and investigation phase. Teams whose on-call burden is primarily correlation work — gathering signal from multiple observability tools and synthesizing a root cause — rather than decision-making. Organizations looking to reduce MTTR without adding headcount to the rotation.