dartantic_interface 3.0.0

Interface definitions for dartantic_ai for use in building custom providers.

3.0.0

Streaming Thinking via ChatResult.thinking

Added optional thinking field to ChatResult for streaming thinking content. During streaming, thinking deltas are delivered via chunk.thinking for real-time display, while the consolidated ThinkingPart appears in final messages for history.

Breaking Change: Migrated to genai_primitives Types

Core message and part types are now provided by the genai_primitives package and re-exported from dartantic_interface. This provides better interoperability with other GenAI tooling in the Dart ecosystem.

Types re-exported from genai_primitives:

• ChatMessage, ChatMessageRole
• StandardPart, TextPart, DataPart, LinkPart, ThinkingPart
• ToolPart, ToolPartKind
• ToolDefinition

Part Type Alias (genai_primitives 0.2.0 compatibility)

In genai_primitives 0.2.0, the Part type hierarchy was reorganized:

• Part became the extendable base class
• StandardPart is the sealed class containing built-in part types

Dartantic provides Part as a typedef alias for StandardPart, so existing code continues to work unchanged. If you need to create a custom Part that extends the base class, import genai_primitives directly:

import 'package:genai_primitives/genai_primitives.dart' as gaip;
class MyCustomPart extends gaip.Part { ... }

Breaking Change: Migrated to json_schema_builder for Schemas

The Schema type is now provided by the json_schema_builder package. Use the S builder class for constructing schemas:

import 'package:dartantic_interface/dartantic_interface.dart';

// Build schemas with S.*
final schema = S.object(properties: {
  'name': S.string(description: 'User name'),
  'age': S.integer(minimum: 0),
});

New: ThinkingPart for Extended Reasoning

Added ThinkingPart to represent extended thinking/reasoning content from LLMs. This provides a unified representation across providers that support thinking (OpenAI Responses, Anthropic, Google).

New Convenience Extensions

Added MessagePartHelpers extension on Iterable<Part>:

• thinkingParts - extracts all ThinkingPart instances
• thinkingText - concatenates all thinking text

2.0.0

• BREAKING: Removed ProviderCaps enum from the interface. Provider capabilities were only meaningful for testing default models and provided no provider-wide guarantees. Capability filtering for tests is now handled via ProviderTestCaps in dartantic_ai's test infrastructure. Consider the Provider.listModels method for run-time model details, e.g. chat, embedding, media, etc.
• Removed caps field from Provider base class.

1.3.0

• introduced media generation primitives (MediaGenerationModel, MediaGenerationResult, MediaGenerationModelResult, and MediaGenerationModelOptions)
• extended Provider with media factory support and added ProviderCaps.mediaGeneration
• added ModelKind.media for provider defaults and discovery

1.2.0

• added optional 'thinking' field to ChatResult for enhanced reasoning output
• updated Provider to support thinking feature toggle

1.1.0

• ProviderCaps.vision => ProviderCaps.chatVision to tighten the meaning

1.0.5

• made usage nullable for when there is no usage.

1.0.4

• added ProviderCaps.thinking

1.0.3

• remove custom lint dependency

1.0.2

• fixed a compilation error on the web

1.0.1

• downgrading meta for wider compatibility

1.0.0

• initial release