Aurora

Cross-platform AI client built with Flutter

中文

Preview

    

    

About

Local-first cross-platform AI client. Chat, MCP, Skills, knowledge base, translation, and sync in one app.

• Multi-provider, multi-protocol routing with flexible model composition.
• Core data stored locally by default.

Platform Support

Platform	Status	Notes
Windows	✅	Primary desktop platform. Fluent style, window management, tray, desktop Skills.
macOS	✅	Full desktop features with native layout.
Linux	✅	Full desktop features.
Android	✅	Mobile-adapted with attachments and Bottom Sheet interaction.
iOS	✅	Mobile-adapted with attachments and Bottom Sheet interaction.

Features

1. Chat & Session Management

• Locally persisted chat history, session tree, topics, token stats, and attachment references.
• Tree-style history with search, rename, delete, drag-to-reorder, and session branching.
• Streaming output, thinking/reasoning display, message retry, edit, and delete.
• @ to switch models, / to switch presets in the input area.
• Markdown, code blocks, tables, images, footnotes, and LaTeX rendering.
• Message bubbles show TTFT, TPS, token usage, timestamps, and tool output.

2. Multi-Provider & Model Routing

• Built-in OpenAI and custom providers, compatible with OpenAI API, Gemini Native, and Anthropic Messages.
• Model list fetching, capability-based routing, provider/model toggle and sorting.
• Multi API key rotation, request timeout, global/model-level parameter configuration.
• Assign models per capability: chat, embeddings, image, speech, transcription, translation.

3. Assistant System

• Multiple assistants, each with its own avatar, name, description, and system prompt.
• Each assistant can bind independent Skills, MCP servers, and knowledge bases.
• Assistant-level long-term memory with configurable Provider/Model for memory distillation, or follow the current chat model.
• Memory auto-distills user preferences (language, length, format, tone, code style, etc.).

4. MCP & Tool Calling

• Transport: stdio, streamable HTTP.
• Connection testing, status view, reconnect, latency display, tool caching, error output.
• Three-level binding: global, session, assistant.
• Expose MCP tools directly to models during chat.

stdio is desktop-only; mobile exposes HTTP-type MCP servers only.

5. Skills Plugins

• Reads SKILL.md / SKILL_<lang>.md from skills/ to declare local skills as callable tools.
• Tool types: shell, HTTP, API.
• Models can auto-select and execute based on skill manuals.
• Suitable for packaging fixed workflows into reusable plugins.

Shell-type Skills are desktop-only.

6. Knowledge Base (RAG)

• Create local knowledge bases, import files for chunked retrieval.
• Lexical search + optional embeddings search.
• Retrieval results injected into context; models can rewrite the search query.
• Assistants can bind independent knowledge bases.

7. Web Search & Translation

• Built-in web search; results injected into conversation before generating answers.
• Search configurable by engine, region, safe search, result count, timeout.
• Standalone translation page with auto language detection, side-by-side view, one-click copy.

8. Multimodal

• Image, audio, video, and file attachment input.
• With compatible models: image understanding, audio transcription, translation, image generation, etc.

Actual capabilities depend on provider, model, and routing configuration.

9. UI, Settings & Operations

• Bilingual (EN/ZH), light/dark/custom themes, accent color, background image, blur and brightness control.
• Desktop tray, window close behavior configuration, recent session restore.
• Prompt presets with {time}, {user_name}, {system}, {device}, {language}, {clipboard} variables.
• Built-in log viewer, error records, usage statistics, model capability lab.

10. Sync & Backup

• WebDAV remote backup, restore, and management.
• Local export/import.
• Selective backup scope: chat history, presets, provider config, app settings, assistants, knowledge bases, usage stats.

Project Structure

lib/
  core/        Infrastructure, bootstrap, and common utilities
  features/    Chat, assistant, MCP, knowledge base, sync, etc.
  l10n/        Internationalization resources
  shared/      Shared services, themes, components, and tools
packages/
  aurora_search/  Search capability wrapper
skills/
  Skill examples and extension entry point

Limitations

• Shell Skills and stdio MCP are desktop-only; mobile uses HTTP MCP.
• Multimodal capabilities (image, speech, transcription, translation) depend on actual model support.

License

MIT License