Skip to content

SerOps documentation

SerOps is open-source AI root cause analysis, self-hosted on your own infrastructure. It connects to your existing observability, SCM, and CI/CD tools — correlating logs, metrics, distributed traces, code changes, and deployments to automatically detect issues and explain their root cause.

  • Getting Started

    Set up SerOps and create your first project in minutes.

    Quick Start

  • Connect Your Stack

    Bring your own observability, SCM, and CI/CD tools. SerOps integrates with Elasticsearch, Prometheus, Datadog, GitLab, GitHub, and more.

    Integrations

  • AI-Powered Correlation

    Pluggable AI models analyze your logs alongside metrics, distributed traces, and code changes to pinpoint root causes and recommend fixes.

    AI Chat

  • Security First

    All data is scanned for PII and credentials before AI analysis. SSO, RBAC, and full audit logging included.

    Data Protection

How SerOps Works

SerOps sits on top of the tools you already use. It takes in five kinds of signal, prepares them, and uses AI to connect the dots:

How SerOps works: logs, metrics and traces are pulled, code changes and CI/CD pipelines arrive by webhook; preprocessing (dedup, anomaly detection, span summary, PII redaction, cost gating) feeds AI analysis, which produces the root cause with commit correlation, then alerts and escalation. How SerOps works: logs, metrics and traces are pulled, code changes and CI/CD pipelines arrive by webhook; preprocessing (dedup, anomaly detection, span summary, PII redaction, cost gating) feeds AI analysis, which produces the root cause with commit correlation, then alerts and escalation.

Traces are the one signal that can arrive either way: ship them to SerOps directly over OTLP, or leave them where they are and let SerOps query Jaeger, Zipkin or Datadog APM. Nothing reaches an AI provider raw — see preprocessing below.

Example: SerOps detects a spike in database connection errors in your production logs, correlates it with a CPU spike from Prometheus, sees from the traces that the latency is concentrated in one downstream call, and links it to a commit pushed 30 minutes ago — then alerts your team via Slack with a root cause analysis and fix recommendation.

Traces matter most when the logs say nothing. A dependency that slows from 20ms to 900ms still returns HTTP 200 and logs no error; span duration is the only signal that shows it, and the only one that attributes the delay to a specific hop.

What happens before the AI sees anything

Raw signals do not go to the model. Each batch is prepared first, which is what keeps analysis affordable and keeps your data out of a third party's hands:

Step What it does
Presence gate A window with no logs, no metrics and no spans never becomes a batch. Any one of the three is enough to analyse
Log deduplication Repeated lines collapse to one entry with an occurrence count
Signal filtering ERROR / WARNING / CRITICAL lines and error-shaped messages keep their slot in the prompt; a 100-signature cap stops one noisy loop crowding out everything else
Metric preprocessing Per-service statistics and sustained-anomaly detection, rather than raw time series
Trace summarisation Thousands of spans reduce to the slowest and the failing ones, with the services they touched
PII redaction Emails, tokens and secrets are masked before anything leaves for an AI provider, traces included. See Data Protection
Sentinel cost gating A quiet batch, judged against a rolling baseline, skips the AI call entirely

Key Features

Feature Description
Issue Detection AI classifies 26+ issue types with smart fingerprinting and deduplication
Commit Correlation Links production issues to the exact code changes and deployments that caused them
Metric Correlation Detects infrastructure anomalies (CPU, memory) that coincide with application errors
Trace Correlation Attributes latency to a specific service hop from span durations, including incidents that produce no error log at all
Multi-Environment Separate analysis for development, staging, and production with branch pattern matching
AI Chat Conversational AI for investigating issues with full context from logs, metrics, traces, and code
Model Comparison Re-analyse batches with different AI models and compare results side-by-side
PII Redaction Automatic scanning and masking of sensitive data before AI processing
Notifications Slack, Teams, PagerDuty, OpsGenie and email, with routing rules, quiet hours, digests and escalation
Issue Lifecycle Muting, auto-resolution, acknowledgement, and MTTR tracking
Integrations Log, metric and trace backends, SCM and CI/CD providers, GitOps deployment sources — see Integrations. Credentials are encrypted at rest
Team Management RBAC (admin/user/viewer), team invitations, SSO via OpenID Connect
Audit & Compliance Full audit trail with export, data retention policies, and GDPR support

How SerOps compares