What is Herald (formerly RunLLM)?
Herald is an AI-powered Site Reliability Engineering (SRE) tool that proactively detects and investigates potential issues across your entire stack before they impact users. Unlike traditional monitoring tools that rely on static thresholds and pre-written runbooks, Herald builds a dynamic context graph of your observability data, codebase, CI/CD pipelines, documentation, and dependencies. This allows it to understand what "normal" looks like for each data stream and automatically detect anomalies without manual threshold tuning.
When an anomaly is detected, Herald instantly investigates by correlating signals across alerts, tracing, logs, and code changes. It evaluates multiple hypotheses simultaneously, pinpointing root causes with over 70% accuracy on novel incidents—even those never seen before. For example, if HTTP 500 errors spike in a Slack ingestion pipeline, Herald can trace the issue to specific code paths, such as a timeout in the `slack_bolt` library, and provide actionable insights like configuring explicit SDK timeouts.
Key features include:
- **Predictive Detection**: Identifies issues before alerts fire or customers complain, reducing mean time to detection (MTTD).
- **No Runbook Required**: Handles both known and unknown failure modes without pre-written documentation, saving engineering time.
- **Automated Root Cause Analysis**: Investigates across code, infrastructure, and telemetry, providing validated root causes with related error signals.
- **Custom Anomaly Models**: Builds unique models for each data stream, eliminating the need for manual threshold configuration.
Herald is ideal for SRE teams, DevOps engineers, and platform engineers who want to move from reactive firefighting to proactive reliability. It excels in complex microservice environments where traditional monitoring fails due to high cardinality and novel failure patterns. By catching issues early and providing precise root causes, Herald reduces downtime, improves customer experience, and frees up engineers to focus on innovation rather than incident response.
Technical details: Herald integrates with popular observability tools (e.g., Prometheus, Datadog), version control systems (e.g., GitHub, GitLab), and incident management platforms (e.g., PagerDuty, Slack). It uses machine learning for anomaly detection and a graph-based reasoning engine for correlation analysis. The platform is cloud-native and designed for high-scale production environments.
Use Cases
Key Features
AI powered anomaly detection dynamic context graph proactive issue detection multi signal correlation root cause analysis predictive detection no manual threshold tuning actionable insightsWho is it for?
SREs, DevOps engineers, platform teams, incident responders, engineering managers, observability teams
