Shipped · Observability

What langfuse/langfuse shipped

The public repository of Langfuse · github.com/langfuse/langfuse

Written by FoxPlug from the repository's own releases; not affiliated with Langfuse. 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/langfuse/langfuse; Langfuse 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

Or Use it as a GitHub Action

See the update FoxPlug writes for langfuse, free

Week of November 2, 2026

What shipped

Why it matters

This week adds IAM role auth for S3 in two places, so exports and media storage no longer require stored access keys. Several fixes target dashboards and monitors: overflowing chart lists get a scroll cue, far-future end timestamps no longer hang ClickHouse, and alerts now show the metric value that triggered them. Usage estimates from content inference were removed because they can mislead when providers use caching or hidden reasoning.

Changelog entry

Week of October 26, 2026

What shipped

Why it matters

This week the AI Gateway moved from setup work toward day-to-day use: a dedicated membership role, analytics with user and API key filters, invite flows, and requests always recorded on Cloud. Beyond the gateway, self-hosted operators get safer upgrades and fewer failed queries during Postgres restarts, and automations now cover skill changes.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

Langfuse 4.72.0 is out. AI Gateway requests are attributed to their gateway API key and service, a new AI_GATEWAY org role limits access to the gateway alone, and analytics can be filtered by user and API key. Also this week: skill changes can trigger automations, and Prisma read

Post for LinkedIn

Langfuse 4.72.0 is out. AI Gateway spans are now attributed to their gateway API key and service, self-hosted instances report gateway request counts, and PagerDuty can receive alerts, project notifications and prompt changes. The AI Gateway settings gained an analytics tab, and analytics can be filtered by user and API key. A new AI_GATEWAY organization membership role gives a member gateway access without the rest of the organization. Outside the gateway: skill changes can trigger version automations, a dedicated monitor automation dialog is in, and Prisma reads retry on transient Postgres connection errors. ClickHouse migration 0050 now handles legacy scores columns so self-hosted upgrades complete.

Week of October 19, 2026

What shipped

Why it matters

This week adds the first public API surface for Topics, more ways to find and filter material across the app, and clearer first-run and setup screens. It also tightens how stored secrets are handled when destinations change and adds more checks around batch runs and transcripts.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

This week: read-only Topics endpoints under the unstable public API, Cmd+K jumps to prompts, datasets, queues and evaluators, a batch evaluation tool with batch_action_id on outputs, and webhook or PostHog destination changes now require secrets again.

Post for LinkedIn

This week's release work on Langfuse: read-only Topics endpoints land under the unstable public API while Topics stays in closed beta. The Cmd+K menu now jumps to prompts, datasets, annotation queues and evaluators, and a batch evaluation tool adds a batch_action_id on batch run outputs. Changing a webhook or PostHog destination URL requires secrets again so stored write-only secrets are not sent to a new host, and legacy plaintext webhook headers are now masked and encrypted. Trace and user tables gain separate Skill Names Available and Loaded filters, and the first Topics run shows analyzing, waiting and failed screens.

Week of October 12, 2026

What shipped

Why it matters

This week tightens access control on API keys and the public API, and makes query and billing paths more consistent for self-hosted and cloud users. Cost and usage views gain more accurate numbers, and skills move into a public preview so teams can browse and measure them.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

This week: project API key permissions on by default with ingestion read access, public API refuses cross-project keys, topics show run costs, AI Gateway keys split from org keys, soft-deleted rows hidden on events reads, and skills open as a public preview. v4.62.0 and v4.63.0 r

Week of October 5, 2026

What shipped

Why it matters

This week brought a major release with UI and session improvements, plus stronger credential handling for FIPS mode and faster list and dashboard reads through the read replica. It also added worker startup jitter to smooth Redis load and graduated decision-model evaluators out of experimental status. Together these changes affect security, database read performance, and daily evaluation workflows.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

V4.56.0 is out with a design-system CodeSection, syntax highlighting in session tool previews, and ClickHouse spend alerts. Also this week: PBKDF2 password hashing in FIPS mode, read-replica routing for score and dashboard reads, custom table row heights, and worker startup jitte

Post for LinkedIn

Version v4.56.0 shipped this week. It replaces CodeView with a design-system CodeSection, adds syntax highlighting to session tool previews, sends spend alerts for ClickHouse-billed organizations, and more. On the backend, FIPS mode now hashes and verifies passwords with PBKDF2-HMAC-SHA256, new API keys no longer bcrypt-hash their secret, and FIPS mode refuses to start with storage options that compute MD5. List-view observation score reads and v1 dashboard score-widget reads can now go to the read replica, and events list queries disable read-in-order buffering. Workers gained optional startup jitter before queue registration. In the UI, tables with the row-height switch can be dragged to a custom height, and decision-model evaluators are no longer marked experimental.

Week of September 28, 2026

What shipped

Why it matters

This week focused on error handling consistency and evaluation features. Multiple PRs fixed HTTP 500 errors that should have been 404s when operations target missing resources, improving API correctness. Evaluation improvements include decision-model evaluator support, execution health tracking, and rule idle time configuration, making evaluators more flexible and observable.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

V4.50.0 is out. Added decision-model evaluators, seven-day execution health tracking, facet types, and fixed a dozen 500→404 error responses. Error IDs now visible in toasts.

Post for LinkedIn

V4.50.0 released. New evaluation features include decision-model evaluators, seven-day execution health tracking, facet types, and rule idle time configuration. On the API side, a dozen endpoints now correctly return 404 for missing resources instead of 500 errors. Error IDs are now visible in tRPC error toasts and included in issue reports. Tool filters added to session tables.

Week of September 21, 2026

What shipped

Why it matters

This week brings significant UX refinements to experiment comparison and dataset viewing, new skill management capabilities, and multiple performance optimizations for dashboard queries and data retrieval. The changes make core workflows faster and cleaner, with better defaults and improved data presentation throughout the interface.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

V4.46.0 ships cleaner experiment comparisons, formatted dataset JSON views, skill management APIs, and performance improvements across dashboard queries and data retrieval.

Post for LinkedIn

V4.46.0 is out. Experiment side-by-side comparisons now feature collapsible summaries and better score details. Dataset items display formatted JSON for easier viewing. New REST endpoints enable skill management. Performance optimizations route dataset queries to read replicas and fix dashboard query parameter handling. Plus UI improvements throughout including faster tooltips and better breadcrumb styling.

Week of September 14, 2026

What shipped

Why it matters

This week includes structural improvements to core data APIs like threads and gateways, better trace observability with metrics and UI reorganization, and expanded automation capabilities. Infrastructure changes improve event-loop monitoring and batch processing visibility for production deployments.

Changelog entry

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

This week: thread transcripts with conversation history, gateway metadata reorganization, label-based prompt automation filters, and improved trace-level visibility. Full details at the preview links.

Post for LinkedIn

This week's releases focus on better data organization and observability. Threads now return full conversation history per turn, gateway generation metadata is consistently namespaced, and automations support label-based filtering. Infrastructure improvements include event-loop delay metrics for autoscaling and batch processing visibility. Check the preview links for details.

Weeks with too little public activity are left out rather than filled in. Last updated 2026-11-09.

Is this your repo? Ask us to remove this page.