mercuretechnologies/xprem v3.2.6 on GitHub

Xprem 3.2.6

Release date: 2026-10-10

Error tracking (Enterprise)

A new Errors view in Observe groups every JavaScript error your apps report by fingerprint, with occurrence counts, impacted installations and fatal occurrence counts. Each group shows its breakdown by platform, app version, update and device, plus an opt-in histogram. Queries span up to 31 days. The MCP server exposes the same data through two new tools, query_errors
and get_error_details.

Symbolicated stack traces (Enterprise)

eoas publish now ships the Hermes source map of every bundle alongside the update. When the server runs with UPLOAD_SOURCEMAPS=true, it stores the map, indexes it in the background and symbolicates crash stack traces
in the dashboard: original file, line, column and a few lines of source context around each frame. Index jobs are visible from the dashboard and can be rebuilt on demand.

Source maps embed your source code, so they live in their own store.
Point the server at it with the variable matching your storage mode:
S3_BUCKET_SOURCEMAPS_NAME, GCS_BUCKET_SOURCEMAPS_NAME, AZURE_BLOB_SOURCEMAPS_CONTAINER_NAME or LOCAL_SOURCEMAPS_BASE_PATH.
Reusing the updates bucket is allowed only when it is not publicly readable: the server refuses to boot if CDN_BASE_URL fronts it.
Symbolication requires an Enterprise license; the upload itself does not.

Fleet and release insights

The Observe overview shows update adoption per channel and the global breakdown of your fleet by platform, app version and runtime.
The Releases panel counts adoption per channel and runtime version, and only shows the runtime of the newest head.
Update details list the errors reported on that update.
Metric timelines stay readable on sparse data, show releases by default and highlight the latest point.

Update health from the SDK's own signals

Update health no longer depends on the manual xprem_js_crash and app_started events. A fatal exception reported by expo-app-metrics marks the running update as faulty; a successful first render (TTR/TTI) marks it healthy again. The manual events keep working.

MCP page in the dashboard

A new MCP page under the Server section gives connection instructions per client: Claude Code, Claude, Cursor, VS Code, Codex and Gemini CLI.

Ingestion rate limits

Telemetry ingestion is rate limited per address and per app. Defaults are 120 batches per minute from one address to one app, and 6000 per minute for one app in total. Tune or disable them with OBSERVE_INGEST_LIMIT_PER_IP and OBSERVE_INGEST_LIMIT_PER_APP (0 disables a limit).

eoas CLI

eoas publish exports only the native platforms configured in your project. It no longer tries to export web.
Publishing from Windows works: export paths are normalized to forward slashes.
Source maps are emitted by default. Opt out with --no-dumpSourcemap.

Fixes

A malformed expoConfig.json is rejected at publish time instead of producing a broken update. Transient storage errors while reading it are retried, not reported as invalid.
An upload is kept when source map verification fails temporarily.
Publishing finishes server-side even if the CLI disconnects mid-upload.
Manual crash events are fingerprinted at every severity, not only fatal.
Admins can reset another account's password from the Users page.
Role selectors on the Users page stay visible and scrollable on mobile.

Upgrade notes

Set UPLOAD_SOURCEMAPS=true and the matching *_SOURCEMAPS_* variable to enable source map storage. The Helm chart exposes it as uploadSourcemaps. Nothing changes if you leave it unset.
Upgrade eoas to 3.2.6 in your project's devDependencies before upgrading the server.
Database migrations run automatically at boot.

New Contributors

@hhwjsw711 made their first contribution in #270
Full Changelog: v3.2.4...v3.2.6