Seer goes GA, Conversations get a transcript, and AI Guard expands

Seer is generally available — and it works with GitLab

The biggest story of the week is Sentry's Seer leaving beta and becoming generally available to all users. Seer, Sentry's AI debugging agent, can now analyze GitLab repositories as part of its root cause analysis, reading your code files to trace errors back to their source. Once it identifies a fix, Seer can open a Merge Request directly — taking you from error to proposed fix without leaving your workflow. This marks a significant step in making AI-assisted debugging a core part of the error resolution workflow, not just a novelty.

A redesigned Conversations view, smarter issue sorting, and log pinning

Sentry also shipped several UX improvements. The Conversations detail page has been redesigned with two new views: a Transcript view that shows the full user-model-tool interaction in a chat-style layout, and a Timeline view that displays every span in order. Saved conversation lists now support configurable columns and filters. The issue stream now defaults to a "Recommended" sort that combines recency, spikes, and ongoing problems — a meaningful improvement over the old "Last Seen" default. And Log Pinning lets you pin important log lines during debugging, keeping them visible as you investigate and including them in the URL for sharing.

On the SDK side, Sentry JavaScript 10.68.0 added auto-instrumentation for Cloudflare Workers, Durable Objects, and Workflow classes, along with NestJS WebSocket error support in SentryGlobalFilter. The previous 10.67.0 brought SvelteKit 3 support and expanded the orchestrion tracing instrumentation to cover AWS SDK, LangChain, LangGraph, Mongoose, Koa, Remix, and MongoDB. Sentry Python 2.66.1 fixed a critical bug where exceptions raised inside traces_sampler callbacks could crash the application. And Sentry CLI 3.6.2 fixed a subtle config merging issue where auth tokens and URLs from different files could be incorrectly combined.

Datadog: AI Guard expands, Agent gets v3 intake, and Java profiling gets a boost

Datadog's dd-trace-js v6.5.0 and v5.116.0 both added automatic AI Guard evaluation for Anthropic calls, alongside LLM audio capture in messages and OpenAI chat completions. The Datadog Agent 7.81.2 switched to the v3 metrics intake by default, a significant change for metric pipeline performance. The RUM Browser SDK v7.6.1 added debug_ids to profiler traces for better sourcemap unminification and fixed a Nuxt post-hydration error handling bug.

A standout engineering blog post this week details how Datadog worked with SAP, Amazon, and the OpenJDK community to introduce a new CPUTimeSample event in Java Flight Recorder for JDK 25, replacing the biased ExecutionSample event and the need for the unsupported AsyncGetCallTrace API. This is a genuine improvement to Java CPU profiling that benefits the entire ecosystem.

Also notable: dd-trace-js v6.6.0 fixed test optimization to complete runs after subscriber loss, and Sentry CLI 3.6.1 prevented Debug ID collisions between distinct JS files that share identical sourcemaps.

Dash0Operator and other updates

Dash0 Operator 0.150.0 added the Dash0Team custom resource definition and switched Python instrumentation to the Dash0 OpenTelemetry distribution. The Dash0 Node.js distribution received a dependency update. These are smaller releases but show continued investment in the Dash0 Kubernetes-native observability approach.