What openlit/openlit shipped
Written by FoxPlug from the repository's own releases; not affiliated with Openlit. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/openlit/openlit; Openlit did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Microsoft and Okta sign-in can now be configured through environment variables, with the shared groundwork for an instance admin area. Pull request #1896
- Span content now renders as markdown, and JSON values show as a tree instead of raw text, with images blocked and replaced by their alt text. Pull request #1922
- The time window filter gained hour and minute selection so a window can be narrowed to an incident. Pull request #1921
- Span name grouping is restored on the default Traces tab, using the same grouped endpoint as the Spans tab. Pull request #1920
- Text in telemetry list rows can now be selected, so ids and names can be copied. Pull request #1919
- The next and previous arrows in the span detail sheet now keep the selected row's trace instead of leaving the old span on screen. Pull request #1918
- AI Analysis findings are no longer cut off on long traces, and coding-agent cost is no longer double-counted in the dashboard. Pull request #1879
- Session cost in the coding agents hub rollup is no longer double-counted. Pull request #1778
- The CLI now counts each API response once in the session token rollup instead of summing usage over every transcript line. Pull request #1783
Why it matters
This week fixed several day-to-day views in the client: span content is easier to read, traces keeps its grouping, telemetry text can be copied, and span navigation follows the selected row's trace. Sign-in options widened with Microsoft and Okta through environment variables. Cost figures in the coding agent views and the CLI are also more accurate, which matters to anyone reading session totals.
Changelog entry
- Add Microsoft and Okta sign-in configured through environment variables, plus shared groundwork for an instance admin area. Pull request #1896
- Render span content as markdown and show JSON values as a tree instead of raw text. Pull request #1922
- Add hour and minute selection to the time window filter. Pull request #1921
- Restore span name grouping on the default Traces tab. Pull request #1920
- Allow selecting text in telemetry list rows so ids and names can be copied. Pull request #1919
- Keep span detail next and previous arrows on the selected row's trace. Pull request #1918
- Keep AI Analysis findings on long traces and stop double-counting coding-agent cost in the dashboard. Pull request #1879
- Stop double-counting session cost in the coding agents hub rollup. Pull request #1778
- Count each API response once in the session token rollup. Pull request #1783
- Add an opt-in PR validation workflow that runs on request. Pull request #1898
- Add a funding manifest and a well-known manifest URL list. Pull request #1897
Span content renders markdown and JSON as a tree, Traces keeps span name grouping, telemetry text can be selected, and the time window filter now takes hours and minutes. Microsoft and Okta sign-in can be set through environment variables.
Several client fixes landed: span content renders markdown with JSON shown as a tree, span name grouping is back on the default Traces tab, telemetry rows allow text selection so ids and names can be copied, and the span detail arrows follow the selected row's trace. The time window filter now takes hour and minute values. Microsoft and Okta sign-in can be configured through environment variables, alongside groundwork for an instance admin area. Cost reporting is also more accurate: coding-agent cost is no longer double-counted in the dashboard or the hub rollup, and the CLI counts each API response once in the session token rollup.
Week of October 12, 2026
What shipped
- Added a trace-level telemetry view where each row represents one trace or request instead of one span, with filters selecting matching trace ids before loading the complete traces. Pull request #1443
- Added OpenCode integration for coding agents, installing the OpenLIT global plugin in the opencode plugins directory and routing supported calls. Pull request #1591
- Released otel-gpu-collector 1.0.0, which skips write disk metrics for read-only Linux block devices and adds hw.id device attribution to CUDA events on multi-GPU hosts. Release
- Fixed the Together AI instrumentation to bind to together 2.x resource classes, which had been renamed so no Together span was produced. Pull request #1566
- Fixed offline evaluations failing after an OpenLIT restart because the built-in evaluator context blocks were missing from the judge prompt. Pull request #1592
- Fixed operation duration in browser-use and crawl4ai instrumentation to report seconds instead of milliseconds in the span attribute. Pull request #1609
- Fixed GPU process attribution so a process is joined to its own pod by cgroup identity rather than to the pod the GPU was allocated to. Pull request #1719
- Fixed the CLI to report user-facing command errors instead of exiting with status 1 and no output. Pull request #1552
- Added focus rings to shared form controls and fixed dark-mode contrast in the destructive alert variant. Pull request #1564
Why it matters
This week fixed streaming spans that never ended when a caller broke out early or closed a stream, across many provider integrations, and corrected duration units and token accounting that made telemetry misleading. It also added a trace-level telemetry view and OpenCode support, and shipped the 1.0.0 GPU collector with multi-GPU attribution fixes.
Changelog entry
- Added a trace-level telemetry view showing one row per trace or request. Pull request #1443
- Added OpenCode integration for coding agents. Pull request #1591
- Released otel-gpu-collector 1.0.0 with read-only block device handling and hw.id attribution on multi-GPU hosts. Release
- Fixed Together AI instrumentation to bind to together 2.x resource classes. Pull request #1566
- Fixed offline evaluations failing after restart by passing the database config id to the defaults loader. Pull request #1592
- Fixed browser-use and crawl4ai operation duration to report seconds. Pull request #1609
- Fixed GPU process attribution to the process's own pod. Pull request #1719
- Fixed the CLI to report user-facing command errors. Pull request #1552
- Added focus rings to form controls and fixed alert contrast. Pull request #1564
- Fixed VertexAI streaming spans to end on early exit. Pull request #1605
- Fixed Azure AI Inference streaming spans to end on exhaustion and early close. Pull request #1594
New this week: a trace-level telemetry view, OpenCode support, and otel-gpu-collector 1.0.0. Streaming spans now end on early exit across many integrations, and duration and token reporting are corrected.
This week in OpenLIT: a trace-level telemetry view now shows one row per trace or request, OpenCode support was added for coding agents, and otel-gpu-collector 1.0.0 shipped with multi-GPU device attribution. Several fixes correct streaming spans that never ended on early exit or close, operation duration now reported in seconds for browser-use and crawl4ai, reasoning tokens kept out of token usage counts, and the Together integration bound to 2.x classes. Offline evaluations also work after a restart, and the CLI reports user-facing command errors.
Week of September 28, 2026
What shipped
- README, docs, and package metadata now align to position OpenLIT as an open-source agent harness engineering platform for tracing, evaluating, guarding, and improving AI agents on OpenTelemetry. Pull request #1685
- Memcode is now available as an optional backend for OpenLIT's provider-neutral Memory page with encrypted API-key configuration. Pull request #1632
- Login and register screens were redesigned to match the OpenLIT website, removing the two-column feature carousel layout. Pull request #1677
- The Oodle destination documentation now shows how to send both traces and metrics to Oodle from the OpenLIT collector. Pull request #1557
- Fixed 35 broken documentation links in the TypeScript SDK README that were returning 404s after the docs site restructure. Pull request #1678
- Content capture controls are now surfaced in the Python tab of the AI observability quickstart documentation. Pull request #1652
- Removed two dead navigation entries in the docs sidebar that pointed to missing pages. Pull request #1672
Why it matters
This week focuses on aligning OpenLIT's public positioning while fixing documentation gaps and expanding integrations. The repositioning and documentation fixes help new users understand what OpenLIT does, while the new Memcode connector and improved provider attribution strengthen the platform's capabilities.
Changelog entry
- Aligned README, docs, and package metadata to agent harness engineering positioning Pull request #1685
- Added Memcode as optional memory connector backend Pull request #1632
- Redesigned login and register screens; removed feature carousel from auth layout Pull request #1677
- Updated Oodle destination docs to export both traces and metrics to Oodle Pull request #1557
- Fixed 35 broken links in TypeScript SDK README after docs site restructure Pull request #1678
- Added content capture controls documentation to AI observability quickstart Pull request #1652
- Removed dead navigation entries for missing documentation pages Pull request #1672
Week of September 21, 2026
What shipped
- Released openlit-2.1.1. Release
- Replaced the bundled OTel collector with a first-party Go OTLP receiver that writes traces, logs, and metrics to ClickHouse on ports 4318 (HTTP) and 4317 (gRPC). Pull request #1588
- Refreshed the login and sign-up screens with a split layout and added a founder call link in the dashboard header. Pull request #1654
- Added a CE no-op license boot sync hook to support Enterprise licensing without OSS imports. Pull request #1665
- Updated the README with sponsors, deployment partners, and GitHub contributors. Pull request #1651
Why it matters
The removal of the bundled OTel collector in favor of a first-party receiver simplifies the architecture and reduces dependencies while maintaining full OTLP support. The license boot sync hook enables Enterprise features to integrate cleanly with the OSS codebase.
Changelog entry
- Openlit-2.1.1 Release
- Replace bundled OTel collector with first-party Go OTLP receiver for traces, logs, and metrics Pull request #1588
- Add founder call link in dashboard header and refresh login/sign-up screens Pull request #1654
- Add CE no-op license boot sync hook Pull request #1665
- Update README with sponsors, deployment partners, and contributors Pull request #1651
Openlit 2.1.1 is live. We replaced the bundled OTel collector with a first-party Go OTLP receiver and refreshed the auth screens.
Openlit 2.1.1 ships this week with significant infrastructure improvements. We replaced the bundled OpenTelemetry Collector with a first-party Go OTLP receiver that writes traces, logs, and metrics directly to ClickHouse. We also refreshed the authentication experience with a modernized split-layout design and added a direct founder call link in the dashboard.
Week of September 14, 2026
What shipped
- Trustabl scanner connector added with job management, findings tracking, and CLI-driven scan forms. Pull request #1586
- Vertex AI now tracks reasoning tokens from
thoughts_token_countin streaming and non-streaming responses. Pull request #1544 - Streamed tool calls from Groq, AI21, and Together are now captured during streaming. Pull request #1542
Changelog entry
- Add Trustabl scanner connector with GitHub targets, encrypted tokens, CLI-driven forms, and scan job findings Pull request #1586
- Fix Vertex AI reasoning token tracking from
thoughts_token_countin streaming and non-streaming Pull request #1544 - Fix streamed tool calls capture for Groq, AI21, and Together Pull request #1542