# Glassray > Documentation for Glassray, the evaluation layer that makes your AI agents self-improve. - [Welcome to Glassray](https://glassray.ai/docs/getting-started.md): The evaluation layer that makes your AI agents self-improve - learn what your agents are supposed to do, catch recurring deviations in production, and propose code-level fixes. - [Quickstart](https://glassray.ai/docs/quickstart.md): Connect your traces and your code, and let Glassray find recurring deviations and write fixes for your coding agent - end to end. - [Set up with the CLI](https://glassray.ai/docs/cli/setup.md): One command takes your repo from nothing to a watched Glassray account — sign in, connect everything in a quick browser wizard, wire the SDK locally, and confirm a real trace landed before it finishes. - [CLI reference](https://glassray.ai/docs/cli/reference.md): Every glassray command and global flag — the one CLI for cloud setup and local Coach. - [Glassray Coach](https://glassray.ai/docs/coach/overview.md): The local edition of Glassray - run the whole loop on your own machine. One command, one embedded database, zero cloud. - [Coach quickstart](https://glassray.ai/docs/coach/quickstart.md): Get Glassray Coach running, send your first trace, and surface your first deviation - in about three minutes. - [Send traces to Coach](https://glassray.ai/docs/coach/instrument.md): Instrument your agent with the Glassray SDK or any OpenTelemetry exporter and point it at your local Coach. - [Find & fix deviations](https://glassray.ai/docs/coach/analyze.md): The self-healing loop, worked end to end - discover recurring failures, generate a fix for your coding agent, lock the rule in as an eval, and prove you didn't regress. - [From scratch with Claude Code](https://glassray.ai/docs/coach/from-scratch.md): Stand up a brand-new instrumented agent from an empty folder, then let Claude Code find the silent failures and drive the fix - the whole loop, agent-driven, in about fifteen minutes. - [CLI & coding agents](https://glassray.ai/docs/coach/cli.md): Every glassray-coach command, how to choose an analysis model, and how to drive the whole Coach loop from your coding agent with the shipped skill. - [Traces](https://glassray.ai/docs/traces.md): What a trace is and how to get your traces into Glassray - Langfuse, LangSmith, PostHog, Better Stack, Vercel, or OpenTelemetry. - [Flows](https://glassray.ai/docs/flows.md): A flow is one or more agents doing one job - anchored in your code and confirmed by your traces, and the unit Glassray finds deviations against. - [Rules](https://glassray.ai/docs/rules.md): A rule is one expected behaviour of a flow - read from your code, or promoted from a confirmed deviation - and the thing a deviation is found against. - [Deviations](https://glassray.ai/docs/deviations.md): A deviation is a recurring type of misbehavior Glassray finds across your traces - with a review lifecycle, a fix for your coding agent, and linked pull requests. - [Specs](https://glassray.ai/docs/spec.md): A spec is a plain-language description of what your agents are supposed to do - the rubric Glassray finds deviations against. - [Signals](https://glassray.ai/docs/signal.md): A signal is a piece of external feedback about a trace, step, or session - and a negative one can become a deviation. - [Connect Langfuse](https://glassray.ai/docs/langfuse.md): Connect Langfuse so Glassray pulls your traces on a sync cadence. - [Connect LangSmith](https://glassray.ai/docs/langsmith.md): Connect LangSmith so Glassray pulls your traces on a sync cadence. - [Connect PostHog](https://glassray.ai/docs/posthog.md): Connect PostHog so Glassray pulls your LLM observability traces on a sync cadence. - [Connect Better Stack](https://glassray.ai/docs/betterstack.md): Connect Better Stack so Glassray pulls your OpenTelemetry spans on a sync cadence. - [Stream AI traces from Vercel](https://glassray.ai/docs/vercel.md): Connect your Vercel-deployed app to Glassray with a Trace Drain - three pasted values, no code changes to your exporter. - [Send traces with OpenTelemetry (OTLP)](https://glassray.ai/docs/otlp-ingestion.md): Send traces to Glassray from any app that uses OpenTelemetry, with setup examples for Node.js, Python and Ruby. - [SDK quickstart](https://glassray.ai/docs/sdk-quickstart.md): Install @glassray/tracing, wrap your agent, and see your first trace in Glassray - no OpenTelemetry setup required. - [Instrumenting your agent](https://glassray.ai/docs/sdk-instrumenting.md): The three instrumentation modes of @glassray/tracing, the llm/tool/span helpers, sessions, usage + cost capture, and error recording. - [Metadata: customer, agent, flow](https://glassray.ai/docs/sdk-metadata.md): Glassray's metadata convention — the reserved fields with first-class support (customer, agent, flow, userId, version), plus arbitrary custom attributes you can filter traces by. - [Configuration reference](https://glassray.ai/docs/sdk-configuration.md): Every @glassray/tracing constructor option and environment variable — precedence, defaults, and the kill switch. - [Privacy & redaction](https://glassray.ai/docs/sdk-privacy.md): What @glassray/tracing sends, what it never sends, and the four layers of content control - in the order they apply. - [Reliability & serverless](https://glassray.ai/docs/sdk-reliability.md): The fail-open guarantees of @glassray/tracing, the documented queue and retry bounds, and what each serverless platform needs. - [Bring your own OpenTelemetry](https://glassray.ai/docs/sdk-byo-otel.md): Point any OTLP exporter at Glassray instead of using the SDK — and the one delivery contract you must honor: whole trace per request. - [SDK troubleshooting](https://glassray.ai/docs/sdk-troubleshooting.md): No traces appearing? Work down this checklist - key validity, source state, filters, sampling, serverless flushing, and the SDK's own diagnostics. - [GitHub](https://glassray.ai/docs/github.md): Connect your repository so Glassray can read your code to generate a spec - and link the pull requests that fix each deviation. - [Slack](https://glassray.ai/docs/slack.md): Report agent misbehavior to Glassray from Slack - the bot finds matching traces, logs a deviation, and scans for more. - [Connect over MCP](https://glassray.ai/docs/mcp-server.md): Connect Claude Code, Cursor, or Claude Desktop to your Glassray organization and work with your traces, flows, and deviations conversationally. - [Webhooks](https://glassray.ai/docs/webhooks.md): Get a signed HTTP push the moment Glassray confirms a deviation - so your own systems act on it without polling. - [Ask Glassray](https://glassray.ai/docs/ask-glassray.md): Explore your traces in natural language - find a set, analyze it, and refine it across turns. - [Changelog](https://glassray.ai/docs/changelog.md): What's new in Glassray - one curated post a week. The stories on top, the full record below. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.