Traversal

• Letting an Incident Worker join a Sev-1 Slack channel and rule out recent deploys before the on-call engineer arrives
• Routing a noisy alert channel to an Alert Worker so each alert gets a verdict and reasoning
• Asking which change or upstream dependency most likely caused a latency spike across several services
• Calling Traversal from Claude Code over MCP to check a diff against live production before merging
• Generating a cited post-mortem from an incident channel once recovery is verified
• Starting investigations from an internal tool or pipeline through the Sessions API with a service account

Traversal is aimed squarely at large engineering organizations where incidents cross many services and teams, and its documentation is unusually specific about how it gets there: read-only access, a Connector and Processor you control, redaction, and a clear responsibility matrix for SaaS, single-tenant, and BYOC deployments. Workers are the practical differentiator, because an agent that joins the incident channel on its own removes the step of someone remembering to ask. The catch is that nothing about cost is public, so the real evaluation happens in a sales-led pilot. Teams should replay a handful of past incidents during that pilot and check its cited root causes against what actually happened before letting Workers into live channels.

About Traversal

Traversal, a New York company founded in 2023 by researchers in causal machine learning, sits on top of an existing observability stack rather than replacing it. Nothing is installed on your hosts. It connects to tools such as Datadog, Dynatrace, Grafana, Splunk, Prometheus, Elasticsearch, Amazon CloudWatch, Azure, Kubernetes, GitHub, GitLab, PagerDuty, ServiceNow, and Jira, and an optional Connector and Processor that you deploy yourself reach private data sources and reduce and redact data before it leaves your environment. From that data it builds what it calls a Production World Model, a continuously updated map of services, dependencies, baselines, changes, and team knowledge, and a Causal Search Engine tests many root-cause hypotheses in parallel against it. Each investigation returns a root cause analysis with confidence ratings, evidence citations, and recommendations, and can be turned into a post-mortem. You can start investigations from the web app, by @mentioning Traversal in Slack or Microsoft Teams, from the Sessions API, or from Claude Code, Cursor, and other clients through its MCP server, where your local coding agent and Traversal hand next steps back and forth. Traversal Workers remove the need to ask. An Incident Worker (generally available since September 2026) joins an incident channel when it opens, investigates before responders arrive, weighs in on mitigations, verifies recovery against live signals, and drafts the post-mortem. Alert Workers, in public beta, give each alert in a channel a triage verdict, and Custom Workers take a plain-language mission. Access is read-only by default, and Traversal writes only where you grant write scope (or posts its findings in Slack and Teams). Deployment can be multi-tenant SaaS, single-tenant SaaS, or a Traversal-managed BYOC deployment in your own AWS account, with options for your own LLM keys, gateways, or Bedrock. The tradeoffs are price transparency and fit. Pricing is quoted only after a sales walkthrough, investigations are metered in credits, and Traversal itself says smaller environments may not get its full value and that it needs existing observability instrumentation to reason over.