APO (AutoPilot Observability) is an intelligent observability platform designed for engineers and SREs to automate system analysis and troubleshooting. It combines OpenTelemetry and eBPF with LLMs to provide actionable insights, reduce alert fatigue, and streamline root cause analysis, aiming for a 10x cost reduction through efficient data handling and automated workflows.

How It Works

APO utilizes a novel LLM-native data plane that transforms diverse observability data (logs, traces, metrics) into structured anomaly events. These events are correlated with an API-centric service map, enabling precise anomaly detection and cross-domain correlation. Agentic workflows, built with low-code orchestration, allow users to embed their expertise into AI agents for custom automated diagnostics, with built-in workflows for alert validity and root cause analysis. Result verifiability is ensured through visual data charts at each step and cross-validation with eBPF Polaris metrics.

Quick Start & Requirements

• Getting started instructions are available at autopilotobservability.com.
• Comprehensive guides can be found in the documentation.
• Zero-touch tracing agent instrumentation is supported via OneAgent technology for multi-language OpenTelemetry agents.

Highlighted Details

• Agentic Workflows for Observability: Low-code orchestration to create custom AI agents for troubleshooting.
• LLM-native data plane: Transforms data into anomaly events for precise detection and correlation.
• Zero-Touch Tracing Agent Instrumentation: Automatic instrumentation of OpenTelemetry agents.
• Result Verifiability: Visual data charts and cross-validation with eBPF Polaris metrics to combat LLM hallucinations.

Maintenance & Community

• APO is open source and welcomes contributions.
• Community channels include Slack and GitHub.

Licensing & Compatibility

• Licensed under the Apache-2.0 License.
• The open-source and extensible design suggests compatibility with various environments.

Limitations & Caveats

The platform is described as having "state-of-the-art Large Language Models," but specific LLM dependencies, versions, or potential costs associated with their use are not detailed. The "Zero-Touch Tracing Agent Instrumentation" relies on proprietary "OneAgent technology," which may have implications for full open-source integration.