DevIO

DevIO is a privacy-first Flutter chat client for working with local and OpenAI-compatible language models. It gives you a clean cross-platform interface for connecting to Ollama, LM Studio, or any compatible chat-completions endpoint while keeping provider configuration and chat history under your control.

Repository About

Privacy-first Flutter client for local and OpenAI-compatible LLMs, with Ollama and LM Studio support, streaming chat, provider switching, and local chat storage.

Features

• Connect to Ollama, LM Studio, or OpenAI-compatible APIs.
• Stream chat responses from local or self-hosted model servers.
• Switch providers, models, temperature, token limits, and endpoint settings from the app.
• Read LM Studio model metadata for vision, tool-use, reasoning, quantization, context length, and loaded-state indicators.
• Store chat history locally with Drift and SQLite.
• Persist provider settings locally with SharedPreferences and secure storage.
• Attach images for vision-capable LM Studio and OpenAI-compatible models.
• Run across Flutter-supported platforms from one codebase.

Supported Providers

Provider	Default endpoint	Notes
Ollama	http://localhost:11434	Uses Ollama chat and model APIs.
LM Studio	http://localhost:1234	Uses LM Studio native model metadata APIs plus OpenAI-compatible chat completions.
OpenAI-compatible	https://api.openai.com	Works with compatible /v1/models and /v1/chat/completions APIs, including multimodal image content when supported by the model.

Provider implementations live in lib/features/llm/services/providers/.

LM Studio Support

DevIO treats LM Studio as a first-class local provider:

• Discovers models through LM Studio's native /api/v1/models endpoint.
• Falls back to /api/v0/models, then OpenAI-compatible /v1/models, when needed.
• Shows model capabilities from LM Studio metadata instead of relying on hardcoded model-name checks.
• Indicates vision, tool-use, reasoning, format, quantization, parameter size, context length, disk size, and loaded state in the model UI.
• Sends image attachments using OpenAI-compatible multimodal chat content for vision-capable models.

Requirements

• Flutter SDK with Dart 3 support
• Xcode for iOS or macOS builds
• Android Studio or Android SDK for Android builds
• A running model server, such as Ollama or LM Studio

Getting Started

Install dependencies:

flutter pub get

Generate Drift, Freezed, and JSON serialization files when models or database schema change:

dart run build_runner build --delete-conflicting-outputs

Run the app:

flutter run

Run checks:

flutter analyze
flutter test

Environment

DevIO can run without a committed .env file. Local environment values are optional and should stay private.

Create a local .env from the example when you want a default Ollama host:

cp .env.example .env

Example:

OLLAMA_HOST=localhost:11434

The app also stores provider choices, base URLs, API keys, selected models, temperature, and token limits through local preferences.

Project Structure

lib/
  blocs/          Auth state
  cubits/         Chat state
  database/       Drift database and migrations
  features/       LLM, settings, profile, help, feedback, notifications
  models/         Serializable app models
  repositories/   Data access layer
  screens/        App screens
  services/       AI and demo services
  theme/          Material theme
  widgets/        Reusable UI components

Important paths:

• lib/main.dart: app bootstrap, dependency setup, and routing.
• lib/database/app_database.dart: Drift schema.
• lib/repositories/chat_repository.dart: chat persistence.
• lib/features/llm/cubit/llm_cubit.dart: provider configuration state.
• lib/features/llm/services/llm_provider_registry.dart: provider registration.

Development Notes

Regenerate code after changing:

• Drift tables or DAOs
• Freezed state or model classes
• JSON-serializable models

Do not commit local machine state, generated build output, private environment files, IDE workspace settings, dependency symlinks, or downloaded archives.

License

MIT