llm_dart_ai 0.11.0-alpha.1

Framework-neutral AI runtime helpers, text runners, and structured output utilities for llm_dart.

Changelog

[Unreleased]

• Aligned structured output stream events with reference object-stream semantics. streamOutput now emits explicit finish and error events for final parsing, and StreamOutputResult exposes raw text result metadata even when final structured output validation fails.
• Aligned text run result facades with reference step-level projections: generated files, sources, tool calls, tool results, and warnings now aggregate across steps, and generate/stream results expose static and dynamic tool call/result views.
• Refactored structured output into a public facade, output strategy, JSON support, stream result, and runner modules. Public OutputSpec APIs and generateOutput/streamOutput behavior are unchanged.
• Split text call result facades from runner glue while keeping generateTextCall and streamTextCall public behavior unchanged.
• Split StreamTextRunResult and provider cancellation stream support out of the stream text runner implementation without changing stream behavior.
• Split generate text runner support into public facade, tool execution, and prompt replay modules while preserving runner behavior.
• Added an explicit text tool continuation result so generate and stream runners no longer depend on nullable tool execution return values.
• Added a shared text loop continuation helper so stop-condition evaluation and prompt replay are no longer duplicated across generate and stream runners.
• Split generate text result accumulation into content buffering, tool projection, and lifecycle modules while preserving public result collection behavior.
• Split stream text runner lifecycle support into event emission, active run state, and finish/error/abort closure modules while preserving stream behavior.
• Split generate text runner lifecycle support into active run state and finish/error/abort closure modules while preserving non-streaming runner behavior.
• Split structured output runner lifecycle support into final parse/error handling and streaming partial projection modules while preserving output runner behavior.
• Split output spec strategies into output-type-owned modules while preserving public OutputSpec exports and structured output behavior.
• Split output foundation and JSON support into focused type, parsing, validation, value, and diagnostics modules while preserving public exports and output behavior.
• Added shared generateSpeech(...) arguments for outputFormat, instructions, speed, and language.
• Made transcribe(...) require a non-empty mediaType to match the provider-facing transcription contract.

0.11.0-alpha.1 - 2026-05-12

• Alpha release of the shared AI helper package.
• Use this package directly when you want generation helpers without depending on the root llm_dart package.
• Includes text generation/streaming helpers, result accumulation, and structured output utilities.
• Owns the user-facing helper surface over provider contracts, including generateText(...), streamText(...), embed(...), embedMany(...), generateImage(...), generateSpeech(...), and transcribe(...).
• Calls provider-side do* methods internally so app code can stay on stable helpers while provider authors implement lower-level contracts.