Nexus

Nexus is a tool for monitoring AI agents in production and catching silent failures that affect users. It is described as an AI engineer that watches and fixes agents in realtime, with a focus on issues that go beyond basic failure modes.

The product centers on catching issues, analyzing their causes, and tracking them over time. It supports custom failure modes defined in plain English, including cases such as a user rephrasing the agent multiple times, an agent making a choice without referencing available evidence or goals, failed tool calls, user frustration in sessions, agent looping, and information misrepresentation. Nexus also performs automatic root-cause analysis and triage in one continuous flow, pulling context from code, logs, traces, ticketing, and prompts. The page shows example analysis that cross-references issues across patterns so only high-signal problems are surfaced.

It includes performance tracking for failure modes and agent behavior over time, showing how often modes fire and how trajectories change. The page also describes Slack alerts when a silent failure is detected, auto-created Linear tickets with root-cause analysis, logs, and reproduction steps, and a Nexus MCP plug-in for Claude Code or Cursor so work can move directly into fixing with full issue context. Nexus says it drafts pull requests with the changes needed to resolve an issue.

For delivery, the page shows installation with pip using nexus-library and says, Don’t have observability? Just use Nexus. It also offers a Book a Demo link and says it is backed by Character Capital.