Starstruk Logo MarkStarstruk

v1.2.3

roboflow/inferencev1.2.3Apr 17, 2026by PawelPeczek-Roboflow

AI Summary

Introduces nested Workflows (Inner Workflows) for reusability and a new `inference-compiler` extension for ahead-of-time model compilation to TRT engines.

Key Highlights

  • Inner Workflows: nested workflows for reusable sub-pipelines
  • New `inference-compiler` extension for TRT model compilation
  • RF-DETR Segmentation models with NAS support
  • Claude Opus 4-7 added to Anthropic workflow block

New Features

  • Nested Workflows
  • inference-compiler extension
  • RF-DETR Segmentation with NAS
  • Claude Opus 4-7 support

Full Release Notes

# 🚀 Added

## 🧩 Inner Workflows

Workflows can now be **nested inside other Workflows**. A parent Workflow can invoke a child Workflow as a single step, passing inputs through and receiving outputs back — making it much easier to factor large pipelines into reusable, composable pieces.

This unlocks a few patterns that were previously painful:

- **Reusable sub-pipelines** — build a "detect + track + filter" Workflow once and drop it into multiple parent Workflows instead of copy-pasting blocks.
- **Team handoffs** — one team can own and version a child Workflow while other teams consume it as a black box.
- **Cleaner graphs** — complex Workflows stay readable because implementation details hide behind a single step.

That was a major lift delivered by @dkosowski87 💪 

Details: https://github.com/roboflow/inference/pull/2230

## âš¡ `inference-compiler` extension

A new **`inference-compiler`** extension is now available, bringing ahead-of-time model compilation to the Inference runtime. You can now compile models trained on Roboflow platform to TRT engines compatible with your devices.

The compiler is extension to `inference-cli` gated with **Roboflow Enterprise license** - community is encouraged to try it out with public models. For enterprise support - contact Roboflow.

```bash
pip install -U inference-cli
inference enterprise inference-compiler compile-model --model-id rfdetr-nano --api-key XXX
```

Details: https://github.com/roboflow/inference/pull/2186

## 🎭 RF-DETR Segmentation models with NAS

@sergii-bond shipped to `inference` the support for RF-DETR Segmentation models trained on the platform with NAS support.


# 🦺 Maitenance

* Elevate Rafel account to be codeowner by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/2228
* Make roboflow_vision_events@v1 non-SIMD so scalar fields accept dynamic step outputs by @rvirani1 in https://github.com/roboflow/inference/pull/2231
* Fix serverless auth bypass and observability by @hansent in https://github.com/roboflow/inference/pull/2234
* Add claude-opus-4-7 to Anthropic Claude workflow block by @Erol444 in https://github.com/roboflow/inference/pull/2236


**Full Changelog**: https://github.com/roboflow/inference/compare/v1.2.2...v1.2.3