Tracer-Cloud/opensre

OpenSRE – AI‑powered Site Reliability Engineering agents

What it is – OpenSRE is an open‑source framework that lets you run AI agents that help with production incident response. The agents can fetch logs, metrics, traces, runbooks and other operational data from the dozens of tools you already use, reason about the problem, and produce evidence‑backed answers or even trigger remediation actions. It also provides a reinforcement‑learning‑style environment (tests/e2e) for training and evaluating such agents.

Key capabilities

• Tool integration – Connects to 60+ services (LLM providers, observability platforms, Kubernetes, AWS, databases, incident‑management tools, Slack/Telegram, etc.).
• Evidence‑driven workflow – Retrieves relevant context, masks sensitive identifiers, runs a tool‑calling loop, and returns a response linked to the underlying data.
• Multiple interaction modes – Interactive REPL shell, headless CLI (opensre ask …), and an embeddable Python API.
• Customizable deployment – Managed cloud gateway, self‑hosted container, or an EC2 AMI‑based gateway. You can bring any LLM (Anthropic, OpenAI, Ollama, Gemini, Bedrock, etc.).
• Session management – Token‑cost tracking, resumable sessions, and fleet monitoring of local coding agents.
• Training & benchmark – Includes realistic end‑to‑end failure scenarios for reinforcement‑learning research.

Getting started

# macOS / Linux
curl -fsSL https://install.opensre.com | bash -s -- -gh
# Windows (PowerShell)
& ([scriptblock]::Create((irm https://install.opensre.com/install.ps1))) -gh

# Run the CLI (sign‑in required)
opensre

The first launch activates a hosted LLM model. After signing in you can:

• Use the REPL (opensre) to describe incidents in plain language and watch the agent act.
• Run a single turn non‑interactively (opensre ask "why is checkout‑api slow?").
• Embed the agent in Python code via bootstrap.embedded.start_embedded_session().

Typical use‑cases

• Incident triage – Quickly gather correlated logs/metrics, generate a root‑cause hypothesis, and post a summary to Slack or PagerDuty.
• Runbook automation – The agent reads existing runbooks and can execute the prescribed remediation steps.
• Continuous reliability tooling – Integrate the Python API into CI/CD pipelines or monitoring alerts to auto‑diagnose failures.
• Research & training – Use the provided e2e test suite as a benchmark environment for training new AI‑SRE models.

Maturity – Public Alpha (v0.1). Core workflows are usable, but APIs and integrations may still change. CI status badge shows ongoing testing.

License – Apache 2.0.

Where to learn more – Quick‑start guide, full docs, and FAQ are hosted at https://www.opensre.com/docs. A Discord community and a Trendshift ranking page are also linked from the README.