Changelog

Changelog

All notable changes to Flutter AI SDK are documented here.

[1.4.0] - 2026-07-08

💾 Prompt Caching & 📄 Documents

• Prompt caching - AIConfig.promptCaching (PromptCaching, TTL 5 min / 1 h) : cache_control explicite chez Anthropic ; compteurs de cache parsés partout (Usage.cachedTokens, nouveau Usage.cacheWriteTokens)
• Documents - DocumentContent fonctionne sur tous les providers cloud : sources URL chez Anthropic, blocs file base64 chez OpenAI

[1.3.0] - 2026-07-08

🧱 Structured Outputs & 🔢 Token Counting

• Structured outputs - ResponseFormat.json(schema: ...) uses each provider's native guaranteed-schema mechanism (OpenAI json_schema with opt-in strict, Anthropic output_config.format, Gemini responseJsonSchema, Ollama schema format)
• Token counting - FlutterAI.countTokens({message}) and BaseProvider.countTokens: exact server-side counts on Anthropic and Google AI, local estimation for OpenAI/Ollama

[1.2.0] - 2026-07-07

🏗️ Architecture Overhaul & New Features

• Architecture - full restructuring into one-class-per-file modules:
  • config/, models/content/, models/tools/ split up
  • One folder per provider with a dedicated wire-format mapper class
  • BaseProvider owns the shared streaming loop (template method)
  • New ProviderRegistry factory (supports custom provider registration)
• Tool Runner - automatic agentic tool-calling loop (ToolRunner, ExecutableTool): parallel execution, error feedback, iteration budget
• Ollama provider - run local models (Llama, Qwen, Gemma...) with streaming, tools, JSON mode and vision; no API key required
• Anthropic - consecutive same-role messages are merged (required for parallel tool results)
• Dependencies upgraded for Flutter 3.44

[1.1.0] - 2026-07-06

🔄 Model Refresh

• Updated default models to current generations:
  • OpenAI: gpt-5.5 (was gpt-4-turbo)
  • Anthropic: claude-opus-4-8 (was claude-3-5-sonnet-latest)
  • Google AI: gemini-3.5-flash (was gemini-1.5-pro)
• Updated model context limits in ModelContextLimits (GPT-5.x, Claude 4/5, Gemini 3.x)
• Anthropic provider:
  • Never sends temperature and top_p together (rejected by Claude 4+)
  • Maps the refusal and model_context_window_exceeded stop reasons
• Documentation (README + wiki) updated to current model IDs
• Dependencies upgraded (mime 2.x, rxdart 0.28, flutter_lints 6)

[1.0.0] - 2024-XX-XX

🎉 Initial Release

First stable release of Flutter AI SDK!

Features

• Unified API - Single interface for multiple AI providers
• Multiple Providers
  • OpenAI (GPT-4, GPT-3.5, GPT-4-turbo, GPT-4o)
  • Anthropic (Claude 3 Opus, Sonnet, Haiku)
  • Google AI (Gemini Pro, Gemini 1.5)
• Streaming Support - Real-time response streaming with chunk events
• Context Management - Automatic conversation history and memory
• Multimodal Support
  • Text content
  • Image content (URL, bytes, base64)
  • Audio content (OpenAI, Google AI)
  • Document content
• Function Calling - Tool/function support for all providers
• Type Safety - Full Dart type safety with null safety
• Error Handling - Comprehensive error types:
  • AIAuthenticationError
  • AIRateLimitError
  • AIContextLengthError
  • AIContentFilterError
  • AINetworkError
  • AIServerError
  • And more...
• Token Management - Token counting and estimation
• Configuration - Comprehensive configuration options

Documentation

• Complete wiki documentation
• API reference
• Usage examples
• Provider guides

---

Version History Format

Future releases will follow this format:

[X.Y.Z] - YYYY-MM-DD

Added

• New features

Changed

• Changes to existing functionality

Deprecated

• Features to be removed in future versions

Removed

• Removed features

Fixed

• Bug fixes

Security

• Security fixes

---

Versioning

This project follows Semantic Versioning:

• MAJOR (X.0.0): Incompatible API changes
• MINOR (0.X.0): New functionality (backwards compatible)
• PATCH (0.0.X): Bug fixes (backwards compatible)

Upgrade Guide

When upgrading between major versions, check the Migration Guide for breaking changes and how to update your code.