Wisp Science

Summary

Wisp Science is an open-source, local-first AI research workbench that provides a desktop AI assistant and scientific computing environment. It connects to AI models (OpenAI-compatible and Anthropic), runs persistent Python/R environments, and accesses bioinformatics databases via MCP servers while keeping data local.

How It Works

The project uses an agent loop that streams tokens from AI models, executes sandboxed file and shell tools, and leverages persistent Python/R REPLs for interactive computing. It employs an MCP client to discover and use tools from bundled bioinformatics servers (e.g., PubMed) without overloading the model context. Built with Rust (core), Tauri v2 (desktop shell), and Leptos (UI), it stores conversations in a local SQLite database and secures API keys in the OS keyring.

Primary Use Cases

• AI-assisted scientific research (literature search, analysis, computation, writing, task management)
• management)
• Running persistent Python and R environments for data analysis
• Accessing ~80 bioinformatics databases (e.g., PubMed, GEO) via MCP servers
• Creating and using reusable Agent Skills (SKILL.md) for workflow automation
• Managing research projects with local-first data storage and optional encrypted sync

Key Technologies

• Rust (core language)
• Tauri v2 (desktop framework)
• Leptos (UI framework)
• Model Context Protocol (MCP) for tool integration
• SQLite (local storage for conversations and settings)
• OS keyring (securing API keys)
• Python and R (persistent REPLs)