What's Changed

• use api keys in e2e tests by @hansent in #1568
• Add Statefulness to Custom Code Blocks by @yeldarby in #1567
• Propagate remote exception to fix missing metadata in Modal execution error by @brunopicinin in #1565
• handle buffer filling / consumption strategy params when passed as str by @grzegorz-roboflow in #1571
• don't auth /model/registry endpoint via api_key on serverless by @hansent in #1570
• fixing memory leak and variable not associated value on PolygonZoneBlock by @digaobarbosa in #1573
• bump version to 0.56.0 by @hansent in #1574
• PolygonzoneBloc refactor by @digaobarbosa in #1576
• Pass countinference to get model artifacts by @grzegorz-roboflow in #1575

Full Changelog: v0.55.2...v0.56.0

What's Changed

• fix DD and Serverless auth for legacy endpoints by @hansent in #1566

Full Changelog: v0.55.1...v0.55.2

What's Changed

• ⚡️ Speed up function handle_describe_workflows_blocks_request by 84% in PR #1559 (auth-all-serverless-requests) by @codeflash-ai[bot] in #1563
• Move receiving from VideoTransformTrack.recv to independent incoming frames draining loop by @grzegorz-roboflow in #1560
• API key for all requests on servless by @hansent in #1559
• increase timeout INTEGRATION TESTS - Inference Server CPU x86 by @hansent in #1564

Full Changelog: v0.55.0...v0.55.1

What's Changed

• Update sam2-rt model ID in sam2-rt README by @grzegorz-roboflow in #1531
• use Windows-2022 runner for windows build by @hansent in #1534
• asyncua moved User and UserRole definition to asyncua.crypto.permission_rules by @grzegorz-roboflow in #1539
• Fix property definition error when filtering first left-most detection by @grzegorz-roboflow in #1538
• Fix for Dynamic Zone block types by @digaobarbosa in #1533
• Pass countinference and shared_secret to RoboflowInferenceModel by @grzegorz-roboflow in #1546
• fix instance segmentation filter by @digaobarbosa in #1547
• Fix custom block error runtime error metadata by @brunopicinin in #1545
• Wrap state-sensitive models with locks to make the execution of server handler thread safe by @PawelPeczek-Roboflow in #1549
• rlock for owlv2 by @hansent in #1551
• add **kwargs to initialize_model functions that were missing it by @hansent in #1553
• owlv2 lock in DummyOwl class by @hansent in #1554
• Fix modal execution when custom python execution is also allowed by @yeldarby in #1555
• time_in_zone accepting multiple zones by @digaobarbosa in #1542
• Fix modal serialization by @brunopicinin in #1556
• Enable GET /model/registry with feature flag by @grzegorz-roboflow in #1557
• Stabilize webrtc frames quality by @grzegorz-roboflow in #1536
• Feat/custom python blocks modal sandbox by @yeldarby in #1522

Full Changelog: v0.54.2...v0.55.0

🔧 Fixed

• Re-raise model lock acquisition error from Workflows Execution Engine by @PawelPeczek-Roboflow in #1521
• Add changes to secure 'WithFixedSizeCache' from errors related to model acquisition locks by @PawelPeczek-Roboflow in #1526
• Record source when inferring on instance segmentation models by @grzegorz-roboflow in #1525
• Extend Moondream2InferenceRequest to include 'prompt' by @grzegorz-roboflow in #1529

🧑‍🏭 Maintenance

• Gzip workflows schema by @iurisilvio in #1520
• support supervision 0.26.0 by @hansent in #1524
• bump base cpu images from python3.9 -> 3.11 by @hansent in #1528

🥼 inference-experimental changes

• add sam2-real-time to inference_experimental by @grzegorz-roboflow in #1471
• Update uv.lock by @PawelPeczek-Roboflow in #1532

Full Changelog: v0.54.1...v0.54.2

🔧 Fixed

• Fix bug with improper parsing multipart-request causing reading the request body stream twice by @PawelPeczek-Roboflow in #1518

Full Changelog: v0.54.0...v0.54.1

💪 Added

inference server with non-blocking handlers

We redesigned request handling to eliminate event-loop blocking in FastAPI. The server now processes requests non-blocking, enabling higher concurrency and improved throughput while keeping latency about the same.

• 🛠️ Removed event-loop bottlenecks in the request pipeline
• 📈 Higher throughput under load
• ⏱️ Comparable latency

ℹ️ No action required.

💻 Discover server logs with ease

A fresh, at-a-glance dashboard for your Roboflow Inference server—status, metrics, and live logs in one place. All thanks to @digaobarbosa

• 🟢 Health & Status: Server health badge, version, server ID, uptime, and memory usage.
• 📦 Loaded Models: Clear empty state when no models are active.
• 📈 API Requests: Totals and success rate to monitor reliability.
• 🪵 Live Logs: Real-time tailing with level filter, text search, auto-scroll, and “latest” timestamp.
• 🔄 Manual Refresh: Update metrics on demand (Refresh Data).

Where to find it: open the new Inference Dashboard from inference landing page

🔧 Fixed

• Fix copy paste in local builder by @yeldarby in #1513
• Add missing page size parameter to list-ingest-details CLI command by @PawelPeczek-Roboflow in #1516

Full Changelog: v0.53.0...v0.54.0

💪 Added

🌀 Execution Engine v1.6.0

No more wrestling with Batch[T] vs. scalar inputs! The engine now auto-magically wraps scalars into batches when needed, so those annoying compilation errors are history.

• 🪄 Auto Batch Casting: scalars become batches on the fly.

• 📦 Smarter outputs: batches can expand or collapse consistently.

• 👉 Existing workflows keep running just fine. New workflows get all the power-ups. Only some new blocks will need to add additional manifest method to make use of new features

Please review our change log 🥼 which outlines all introduced changes.

🏷️ QR code generation in Workflows

Ever struggled to generate QR code? We have you covered now with Roboflow Workflows ✨ thanks to @Grantimus9. Visit 📖 docs to discover new feature.

🎨 Add icons to your visualisations

Bored with textual labels in your Computer Vision demos? From now on, thanks to @yeldarby you can boost your visualisations with icons. Visit 📖 docs to learn more.

⚙️ Maintanence

• Claude Opus 4 and Claude Sonnet 4 added into roboflow_core/anthropic_claude@v1 block thanks to @brunopicinin in #1496
• Refreshed model versions for Google Gemini block thanks to @brunopicinin in #1495
• Enterprise/Industrial Integration Block Visibility by @NVergunst-ROBO in #1502

🔧 Fixed

• Add env variable to perform atomic cache writes by @sberan in #1503
• Support max detections for OWLv2 models by @yeldarby in #1508
• Add model-instance level thread lock on accessing ONNX session to avoid GPU execution concurrency issues by @PawelPeczek-Roboflow in #1510

🏅 New Contributors

• @Grantimus9 made their first contribution in #1500
• @NVergunst-ROBO made their first contribution in #1502

Full Changelog: v0.52.1...v0.53.0

Inference

• Fault tolerant redis calls for model monitoring by @rvirani1 in #1482
• use macos-14 for app bundle builder by @hansent in #1483
• Update moondream2 endpoint by @grzegorz-roboflow in #1487
• macos app inference improvements by @digaobarbosa in #1488
• Fix nvidia T4 integration tests for moondream2 by @grzegorz-roboflow in #1490
• Pin dockerfiles built on top of Python 3.9 to bookworm due to recent debian trixie release by @grzegorz-roboflow in #1491
• Fix invalid batch_size string value in model registry response by @bigbitbus in #1486
• Fix for code scanning alert no. 156: Workflow does not contain permissions by @grzegorz-roboflow in #1484
• dashboard page by @digaobarbosa in #1489
• Add x-goog-hash MD5 verification if MD5 is available in the header by @grzegorz-roboflow in #1492
• Enable moondream CI by @grzegorz-roboflow in #1493

New Contributors

• @rvirani1 made their first contribution in #1482

Full Changelog: v0.52.0...v0.52.1

Inference:

• Add line number and function name information to dynamic block errors by @brunopicinin in #1468
• When version number is aa.bb.cc, script was not removing leading 'v' by @grzegorz-roboflow in #1470
• Record exec time in serverless v2 when workflow results in exception by @grzegorz-roboflow in #1476
• Removed noisy debug logline by @bigbitbus in #1478
• Gpt 5 OpenAI block by @hansent in #1479
• Install page with download the latest buttons by @digaobarbosa in #1475
• Remove metlo waf by @bigbitbus in #1480

Inference experimental:

• Run uv sync to fix inference-exp CI by @grzegorz-roboflow in #1469
• publish inference experimental to pypi by @grzegorz-roboflow in #1472
• Fix/inference exp pypi workflow filename by @grzegorz-roboflow in #1473
• Lora for remaining transformer models to inference_exp by @hansent in #1462
• Inference exp pypi publisher ci improvements by @grzegorz-roboflow in #1474
• Fix test_clip_onnx_image_prediction_for_numpy by @grzegorz-roboflow in #1477
• Moondream2 for inference_exp by @hansent in #1466

New Contributors

• @digaobarbosa made their first contribution in #1475

Full Changelog: v0.51.10...v0.52.0