Skip to main content
Glassray is in private alpha. Join the alpha and we’ll connect your traces, walk through your agent’s code, and get the loop running on your production data.

What is Glassray?

Glassray is the evaluation layer that makes your AI agents self-improve. It connects to the traces your agents already emit, learns what each part of your system is supposed to do, then scans your production traffic for recurring misbehavior and proposes code-level fixes. Most evaluation tooling scores final outputs. Glassray works from a model of intent: it reads your agent’s code to learn what each step should do, generates a plain-language spec, and checks production traces against it. That’s how it surfaces silent failures - answers that look right but were produced the wrong way.

Learns your system

Reads your traces and your code to map your agents into flows and generate a spec of intended behavior.

Finds recurring deviations

Clusters production traces into recurring types of misbehavior, not just one-off errors.

Proposes code-level fixes

On an accepted deviation, suggests a prompt or logic change as a diff, with the failing trace attached.

Scans for more instances

Runs a deep search across your trace history to find every other place the same deviation shows up.

Who it’s for

Glassray is built for teams running LLM agents in production: multi-step pipelines, tool-calling agents, retrieval chains, and multi-agent systems. If your agent’s traces land in a tracing system like Langfuse, LangSmith, or PostHog - or you can push OpenTelemetry - Glassray can work with them.

Next steps

Quickstart

Connect your data and run the loop, step by step.

Flows

How Glassray groups your traces into flows - the unit it tracks quality on.

Connect over MCP

Let your coding agent query your org’s traces, flows, and deviations.

Join the alpha

Get access and we’ll get the loop running on your production data.