xuzhougeng/wisp-science

What it solves

Wisp Science provides a local-first, integrated workbench for AI-assisted scientific research. It eliminates the fragmentation of research workflows by combining literature search, code execution (Python and R), database querying, and project documentation into a single environment where data and credentials remain on the user's machine.

How it works

The platform uses an agentic system where users can bring their own LLM (OpenAI-compatible or Anthropic) to drive the research process. The agent can read/write project files, execute shell commands, and utilize reusable "Skills" to perform tasks. It integrates with approximately 80 scientific databases via MCP servers and supports persistent Python and R kernels across different conversations. Compute can be handled locally, via WSL, or on remote SSH hosts.

Who it’s for

It is designed for scientists and researchers who need a secure, local-first environment to manage complex data analysis trajectories, query scientific databases, and maintain a verifiable trail of evidence for publications.

Highlights

• Local-First Privacy: Data, conversations, and credentials stay on the user's machine.
• Scientific Integration: Bundled MCP servers for PubMed, GEO, and ~80 other scientific databases.
• Flexible Compute: Support for local, WSL, and remote SSH hosts with isolated kernels per conversation.
• Research Management: Features like "Explorations" for branching research directions and "Evidence Capsules" for freezing manuscript revisions.
• Extensible Agent: Support for custom Skills (SKILL.md) and approval gates for agent actions.