flutter_agentic 2.0.0

Build AI agents in Flutter — ReAct loop, 7 providers, on-device LLMs, per-call routing, and multi-step flows.

Changelog

All notable changes to flutter_agentic will be documented in this file.

The format follows Keep a Changelog. This project adheres to Semantic Versioning.

---

2.0.0 — 2026-07-02

Changed — Breaking

• All public classes renamed from Genesis* to Agentic* for consistency with the package name:
  • GenesisAgent → AgenticAgent
  • GenesisAI → AgenticAI
  • GenesisHub → AgenticHub
  • GenesisTools → AgenticTools
  • GenesisTool → AgenticTool
  • GenesisFlow → AgenticFlow
  • GenesisLogger → AgenticLogger
  • GenesisModels → AgenticModels
  • GenesisAIException → AgenticException
  • GenesisHubPlatformPaths → AgenticHubPlatformPaths

---

1.0.0 — 2026-06-12

Changed

• Package renamed from genesis_ai_sdk to flutter_agentic — new import: import 'package:flutter_agentic/flutter_agentic.dart';
• Version bumped to 1.0.0 to reflect production-ready stability (311 tests, 160/160 pub.dev score)
• Companion packages now published: flutter_agentic_graph, flutter_agentic_ui, flutter_agentic_tools, flutter_agentic_memory

---

0.2.0 — 2026-06-06

Added

Per-Call Routing

• PolicyRouter — routes every call independently across any number of providers using ordered, user-defined rules. The call site never changes; each task is served by whichever provider the rules choose (cheap/private tasks on-device, hard reasoning in the cloud).
• RouteRules — ready-made rules: sensitive() (PII keywords stay local), shortInput(), longContext(), needsTools(), streaming(), custom().
• RouteContext — per-call info exposed to rules (messages, tools, latest user message, total context length, streaming flag).
• RouteDecision + onRoute callback — observe every routing decision for logging, cost dashboards, or a "running locally 🔒 / cloud ☁️" UI badge.
• Automatic fallback to the default provider when a routed provider fails.
• AgenticAgent.chat() / chatStream() now accept an optional provider override to force a specific provider for that call only.

Flows (Genkit-inspired)

• AgenticFlow — chain multi-step AI pipelines with named, typed, observable steps: AgenticFlow.start<String>('trip-planner').then(...).map(...).
• FlowContext — shared state between steps (ctx.set() / ctx.get()).
• FlowEvent (FlowStepStarted / FlowStepCompleted / FlowStepFailed) — live progress events for step-by-step UI.
• FlowException — failures carry the flow + step name for fast debugging.

Improved

• SmartRouter.stream() now falls back to the secondary provider when the primary stream fails before emitting output (previously: no fallback).

---

0.1.1 — 2026-05-29

Fixed

• Shortened pubspec description to meet pub.dev requirements (improves pub score from 150 to 160)

---

0.1.0 — 2026-05-25

Added

Core

• AgenticAgent — unified agent API with chat(), stream(), tool calling, and multi-turn memory.
• AgenticHub — one-stop factory: fromHuggingFace(), fromUrl(), fromOllama(), fromHFCloud(), fromFile(), fromProvider().
• AgenticHubPlatformPaths.platformModelsDir() — returns the correct writable model directory on every platform (uses path_provider).

Providers

• GemmaProvider — on-device inference via flutter_gemma 0.16.x; supports .litertlm (all platforms), .task (mobile), .tflite, .bin.
• LlamaCppProvider — on-device GGUF inference via llama_cpp_dart 0.2.x; supports macOS, Windows, Linux, Android.
• HFInferenceProvider — cloud inference via the HF Inference Router with multi-backend support: featherless (default), nebius, together, sambanova, hfNative.
• OllamaProvider — local Ollama server with pull(), listModels(), and checkStatus().
• GeminiProvider — Google Gemini API (cloud).
• OpenAIProvider — OpenAI API (cloud, any OpenAI-compatible endpoint).
• AnthropicProvider — Anthropic Claude API (cloud).

Hub / Model Management

• ModelFormat — enum covering all formats: litertlm, task, gguf, tflite, binary, safetensors, onnx, ollama, hfInference, with auto-detection via ModelFormat.detect().
• HFHub — HuggingFace Hub client: modelInfo(), listFiles(), downloadUrl(), parseUrl(), inferenceUrl().
• UniversalModelManager — downloadFromUrl(), downloadFromHF(), pullOllamaModel(), providerForFile(), isDownloaded(), deleteModel().

Tools & ReAct

• AgenticTool / ToolParam — define callable tools with typed parameters.
• ReAct loop with configurable max steps and graceful fallback.

Memory

• InMemoryStore — lightweight in-process message history.
• HiveMemoryStore — persistent cross-session memory backed by Hive.

Safety

• InputGuard — block / rewrite prompts before they reach the model.
• OutputGuard — validate / sanitize model responses.
• RateLimiter — token-bucket rate limiting.
• ConcurrencyLimiter — cap parallel requests.

Routing

• SmartRouter — latency/quality-based provider selection.
• PrivacyRouter — route sensitive prompts to on-device providers automatically.

Platform

• Full platform support table (litertlm on all platforms, task on mobile only, gguf on all except iOS pending xcframework build).
• PLATFORM_SETUP.md — step-by-step setup guide for macOS, iOS, Android, Windows, Linux.