Qurio - Multi-Provider AI Knowledge Workspace

Not another chat UI. Qurio is a research-grade cockpit for real work.

Qurio is a high-velocity AI knowledge workspace built for teams that demand more than basic chat. It supports generic OpenAI-compatible providers, Gemini, SiliconFlow, Kimi (Moonshot), MiniMax, GLM (Zhipu), NVIDIA NIM, and more. Highlights include Deep Research for complex tasks, Custom Agents for specialized workflows, rich tool orchestration (MCP + HTTP), and structured reasoning views with export-ready outputs. The stack is designed around separation of concerns: lightweight models for routing/planning, stronger models for answers, and structured storage for memory and documents.

In one line: plan, tool, and deliver from a single workspace without switching apps.

Screenshots

Home

Conversation + Tool Calls

Deep Research Workflow

MCP Tool Groups

Why Qurio

• Multi-Provider Powerhouse: Connect to almost any model (Gemini, OpenAI, Claude via compatible endpoints, SiliconFlow, Kimi, MiniMax, GLM, NIM, etc.) with dual model slots (Lite + Default).
• Tool Orchestration Beyond LLM Chat: Built-in MCP + HTTP tooling, tool grouping, and per-tool enable/disable controls so agents only see what you want them to use.
• Deep Research, Not Just Search: Dedicated research workflow that plans, executes, and reports with structured steps, citations, and export-ready outputs.
• Agent System That Scales: Per-agent prompts, model preferences, and toolsets with clean switching and workspace binding.
• Document Intelligence: Parse, embed, and retrieve across uploaded documents with hybrid search (vector + keyword) for grounded answers.
• Long-Term Memory That Stays Useful: Domain-indexed summaries with Lite routing so only relevant preferences/background get injected.
• Premium UX: Streaming, rich Markdown/Code rendering, inline images, reasoning visibility, and thoughtful UI for research/think/tool panels.
• Production-Grade: Light/Dark/System themes, secure API key handling, i18n (en/zh), responsive design, Supabase persistence.

Design Philosophy

• Route first, retrieve second: Lite models decide if memory/docs are needed; only then inject relevant snippets.
• Separation of models: Fast models handle planning/routing; stronger models handle final answers.
• Structured memory over keyword lists: Domain-indexed summaries keep personalization usable and controllable.
• Hybrid retrieval: Vector + keyword search balances recall with precision for real-world docs.
• Tool governance: MCP groups, HTTP tools, and per-tool visibility keep agents safe and focused.

Highlights

📱 Mobile PWA Support

• Installable: Add Qurio to your home screen on iOS/Android for an app-like experience.
• Responsive UX: Optimized layouts for mobile, tablets, and desktops with touch-friendly controls.
• Seamless Sessions: Stay signed in and continue where you left off with persistent conversations.

🔬 Academic Research Reports

... View full README on GitHub