Technical overview

Nxolaryn is a runtime recovery and decision-record layer for AI-agent workflows.

This page gives a lightweight technical overview without turning the homepage into documentation.

Deployment assumption

Customer-controlled environments and framework-aware runtime policy hooks.

Primary signals

Repeated steps, invalid tool arguments, lost state handoffs, timeout patterns, and token-spend spikes.

Primary responses

Pause, stop, limited retry, fallback route, schema review, engineering review, or human approval gate.

Policy shape

A typical policy defines what failure was detected, what recovery action is approved, and what non-sensitive event record should be preserved.

Detect

Loop, schema, state, latency, token, or approval issue

Recover

Pause, stop, fallback, limited retry, or human review

Record

Trigger, policy result, action taken, and non-sensitive metadata

Data boundaries

Public contact should not include production credentials, API keys, access tokens, PHI, payment data, customer records, regulated data, raw prompts, raw logs, databases, source code, or sensitive production payloads.

View evaluation boundary