LoginSign Up
← Back to all sparks
B

Bugsnag

INFRA · APIS
Velocity2.5

BugSnag's centre of gravity has shifted from crash reports to performance spans and system metrics.

performance-monitoringmobile-sdksfluttermcpobservabilitymonthly-digest
◆Current state
BugSnag ships one dated digest a month, and for the past six the bulk of the content has been Performance, not error monitoring: span groups, session windows, pre-main iOS start times, and CPU/memory/rendering metrics. Error monitoring still gets work, but it reads as maintenance — filter operators, App Hang detection, HTTP error capture — while the new surface area lands in the Performance product. Flutter has been pulled up to iOS/Android parity over two consecutive releases. A self-hosted MCP server sits alongside all of it, gaining OAuth and write actions rather than staying a read-only lookup.
◆Where it's heading
The product is converging on a single app-health telemetry store where an error, a span, a session, and a device metric are all queryable against each other — Correlated Events in April was the clearest statement of that, and July's per-session aggregated CPU and memory extends it to resource cost. The SDK is becoming the control surface: developers define their own session windows and delivery strategies rather than accepting BugSnag's defaults. The MCP work is the quieter arc, moving from token-based reads to OAuth logins and mutating actions like snoozing errors and linking Jira issues.
◆Prediction
Expect the Sessions tab to gain the same correlation treatment errors got — jumping from a session window straight to the events and spans inside it. The MCP server is the thing to watch: it has been given auth and write actions in consecutive digests, which points at more triage operations exposed to agents next.

◆Recent moves

  1. 8d ago

    SDK-defined session windows get aggregated CPU and memory metrics

    Developers can now carve their own session windows through the SDK API and read mean, min and max CPU and memory for each one, surfaced under a new Sessions tab in Performance. This extends the system-metrics work from fixed app lifecycle events to arbitrary developer-defined flows, so the resource cost of a checkout or a sync can be measured directly. It fits the pattern of handing SDK users the definitions rather than shipping fixed instrumentation.

    View source ↗
  2. 1mo ago

    Side-by-side span comparison lands in Performance

    Comparing two span instances from the same group side by side is the kind of diffing tool that only matters once teams have enough span data to investigate regressions, which marks how far the Performance product has matured. The rest is upkeep: a CLI path fix, Vue Router v5 support in the JS performance SDK, and hardened Cocoa binaries. A thin month against the surrounding releases.

    View source ↗
  3. 2mo ago

    Pre-main iOS start view, MCP server OAuth, Flutter system metrics

    Three strands of the same arc in one digest: pre-main aggregate views for iOS app starts push Performance into startup time that was previously hard to attribute, Flutter system metrics close the gap with iOS and Android, and the self-hosted MCP server swaps manual API tokens for browser-based OAuth. The Flutter parity work in particular removes the caveat that cross-platform apps were second-class in BugSnag's metrics story.

    View source ↗
  4. 3mo ago

    Correlated Events links errors by user, device, session and trace

    Correlated Events surfaces related events sharing a user, device, session or trace directly from the event view — the first feature that treats BugSnag's separate data types as one queryable graph rather than parallel products. Flutter CPU, memory and frame metrics arrive here, a release before the parity claim in May. The 'does not equal' release filter is small but applies across error views, performance views and advanced search alike.

    View source ↗
  5. 3mo ago

    Android App Hang detection and HTTP request error reporting

    App Hang detection on Android and HTTP request error reporting in the Android and Cocoa SDKs widen what counts as a reportable failure beyond crashes and handled exceptions. The supporting work is integration plumbing — short-lived tokens for the Linear plugin, Microsoft Teams webhooks, commenting in the Data Access API — which is the error-monitoring side of the product in maintenance while Performance takes the new capability.

    View source ↗
  6. 4mo ago

    MCP server gains snooze and Jira-linking actions

    Adding tools for snoozing errors and linking Jira issues moves the MCP server past lookups into changing triage state, three months before it gets OAuth in May. The rest is housekeeping: the @bugsnag/source-maps package is deprecated in favour of the BugSnag CLI, and the dashboard warns on links originating in customer-generated content. Worth reading as the start of the agent-facing arc rather than a normal month.

    View source ↗