genkit_llamadart 1.4.0

Genkit Dart plugin for running local models through llamadart.

1.4.0 - 2026-07-27

• Add compatibility with Genkit 0.15.x while retaining the 0.13.2 lower bound.
• Reject new runtime work once plugin disposal starts and dispose independent model runtimes concurrently while preserving every disposal failure.
• Infer media types from URI paths, including signed HTTP(S) image URLs on backends that support remote images, and accept Windows drive-letter media paths.
• Reject unsupported primitive and list tool-input schemas before advertising a mismatched zero-argument tool, while resolving valid reference-wrapped object schemas emitted by Genkit.
• Exercise latest-compatible dependencies in the full CI job and lower-bound dependencies in a focused compatibility job.
• Run every real-model-tagged test in CI and release verification using serialized, immutable, SHA-256-verified Hugging Face fixtures without stealing stale cache locks.
• Correct source and agent example behavior, and refresh lifecycle, capability, architecture, testing, and advanced configuration documentation.

1.3.3 - 2026-06-10

• Widen the Genkit and Schemantic dependency constraints to allow genkit 0.14.x while keeping compatibility with 0.13.x.
• Raise the llamadart dependency floor to 0.8.0 so the package depends on the Dart-only top-level llamadart package metadata.
• Remove the unnecessary Flutter SDK environment constraint from the package metadata.
• Document llamadart Flutter Apple SwiftPM companion packages for iOS and macOS apps.

1.3.2 - 2026-06-07

• Add explicit pub.dev platform metadata for Android, iOS, Linux, macOS, web, and Windows so the package listing reflects the cross-platform support inherited from llamadart.

1.3.1 - 2026-06-06

• Update the llamadart dependency constraint to ^0.7.1 so Genkit models can use .litertlm LiteRT-LM bundles on supported llamadart targets.
• Re-export LiteRT-LM and advanced ModelParams option enums from the top-level package API.
• Clarify README and API docs that model paths are not GGUF-only, and that this release resolves through the Flutter SDK required by hosted llamadart.

1.3.0 - 2026-06-04

• Add plugin and prepared-model APIs for cancelling active llamadart generations without holding the underlying LlamaEngine.

1.2.0 - 2026-05-25

• Add source-backed model preparation with ModelSource, package-managed cache/download options, optional mmproj resolution, and typed prepared-model refs.
• Add observable model preparation tasks with progress snapshots, cache-hit states, failure reporting, and cancellation.
• Add prepared-model createGenkit, warm-up, and ownership-aware disposal helpers.
• Document direct GenUI and genkit_shelf integration paths for prepared local models.

1.1.1

• Update the Genkit dependency constraint to ^0.13.2 and verify compatibility with llamadart 0.6.14.

1.1.0

• Expand README with installation, lifecycle, embeddings, JSON output, and multimodal guidance
• Add standalone embedding and constrained JSON example programs
• Add explicit ai.shutdown() cleanup to example apps
• Add per-model capability flags for embeddings, tools, and constrained output
• Harden streamed tool-call accumulation and raw response metadata
• Add CI and failure-path coverage for plugin, runtime, model, and embedder behavior
• Reorganize source into api, core, and Genkit integration layers
• Mirror the test tree to the source layout with one test per file
• Add optional Hugging Face-backed real-model smoke tests for generation and embeddings
• Add architecture and contributing guides for maintainers and contributors
• Cache downloaded real-model GGUF artifacts in CI

1.0.0

• Add an in-process Genkit plugin backed by llamadart
• Support local path-based chat generation, streaming, embeddings, and tool loops
• Add example apps for basic generation and a multi-turn agent workflow
• Add tests for converters, runtime queueing, embeddings, and Genkit tool loops
• Add pub.dev-ready package metadata and MIT license