NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
ONNX Runtime for Dart — build-time binary via native-assets hook, generalised OnnxSession API, and model-download infrastructure.
Last release 1 months ago
24 Aug 2026
Too new to tell
only 2 dated releases
Nearly every release is documented
notes for 1 of 1 stable releases
Nothing withdrawn
no release was ever pulled
3 months old
2 releases · first in 2026
One column per month.
android_test and ios_test are developer-run and are the only checks of real ORT load and inference on mobile. Nothing enforces that they ran before a
android_test and ios_test are developer-run and are the only checks of real
ORT load and inference on mobile. Nothing enforces that they ran before a
release.
Ranked below the desktop gap: a human does verify mobile each release, so the
exposure is a forgotten step rather than a blind platform. Notes that the
recorded reason for excluding iOS ("cannot run in CI agent") is worth
re-examining, since GitHub-hosted macOS runners ship Xcode and have the
network access an SPM fetch needs.
Co-Authored-By: Claude Opus 5 noreply@anthropic.com
First stable release.
OnnxRuntime.load() failed on macOS Flutter apps with
Failed to load dynamic library 'onnxruntime1220.framework/onnxruntime1220'.
The framework name was derived by stripping every dot from the ORT version,
but Flutter bundles the dylib as onnxruntime.1.22.0.framework. Both
spellings are now probed, so the same build works across Flutter versions.^3.13.0.packages/betto_onnxrt/.Reverted out of the monorepo pubspec for now; doc updates for release
Reverted out of the monorepo pubspec for now; doc updates for release
Initial development release.
hook/build.dart) that downloads and stages the
ONNX Runtime prebuilt binary (v1.22.0) for macOS, Linux, Windows, Android, and
iOS.OnnxRuntime — opens the staged ORT library and initialises the OrtApi
vtable.OnnxSession — generalised FFI inference session supporting arbitrary named
inputs and outputs; exposes output tensor shape and element type.OnnxTensor — typed multi-dimensional array with named factories for
float32, float64, int32, int64, and uint8 element types.SessionOptions — thread-pool sizing for intra-op and inter-op parallelism.ModelDownloader — SHA-256 verified, crash-safe download of ONNX model files
described by a ModelSpec.AllowlistProvider — interface for gating which models ModelDownloader is
permitted to fetch.example/magika/ — a standalone Dart CLI tool that detects file types
using Google's Magika v3.3 ONNX model. Demonstrates end-to-end use of
OnnxRuntime, OnnxSession, and ModelDownloader with a real-world model.
Output mirrors the Python Magika --json format. Run with
dart run bin/magika.dart <file> or compile with dart build cli from inside
example/magika/.Your coding agent can read these notes before it upgrades. Set up the MCP server →