Persistent Python
A long-running local kernel keeps variables, DataFrames, and plots available across agent turns.
Open source · Local-first · Windows / macOS beta
Wisp Science
A local agent workspace for rigorous scientific computing.
Wisp Science runs analysis, database search, Python, MCP tools, and reproducible reports from a desktop app or headless CLI. Your datasets, sessions, artifacts, and API keys stay anchored to your machine.
MCP bio-tools
Connects the agent to PubMed, UniProt, ChEMBL, PubChem, PDB, and other bundled biology database clients.
Traceable artifacts
Tables, code, equations, files, and generated outputs are extracted into an artifact panel tied to the conversation.
RESEARCH WORKBENCH
Built around the pieces scientific agents actually need.
The app combines a model provider, local tools, Python execution, science workflows, and persistent project state instead of acting as a detached chat window.
Persistent Python REPL
A uv-managed kernel worker keeps state alive for iterative analysis.
Biology database tools
Bundled bio MCP servers expose roughly 80 database clients through first-class tools.
29 SKILL workflows
Reusable workflows cover literature review, protein structure, single-cell, chemistry, figures, and remote compute.
Local SQLite history
Projects, messages, frames, settings, and artifacts persist locally; desktop API keys use the OS keyring.
Filesystem and shell tools
Read, write, edit, search, grep, and shell tools run inside the local project boundary.
Context compaction
Three-tier compaction keeps long sessions moving under large context budgets.
RESEARCH FLOWS
Use one session for literature, analysis, and figures.
Wisp Science is strongest when a question needs papers, data files, code execution, and a reproducible trail in the same workspace.
Single-cell analysis
Run Scanpy or scVI-style workflows, annotate cell types, and export UMAPs, marker tables, and methods notes.
Protein structure
Fetch sequences and structures, combine AlphaFold/Boltz/OpenFold-style skills, and draft structured interpretation.
Cheminformatics
Search ChEMBL or PubChem, compare activity data, compute properties, and prepare SAR-style tables.
Literature and writing
Search PubMed or Semantic Scholar, inspect PDFs, draft discussion sections, and keep citations with the generated text.
BUILT-IN DEMOS
Open a read-only session before spending API credits.
Bundled demo sessions show full agent traces and extracted outputs for common life-science tasks.
CRISPR Screen
Raw counts to hit calling and visualization report.
Enzyme Engineering
Sequence analysis, mutation suggestions, and activity reasoning.
Extremophile
Metagenomic search and functional annotation across databases.
Immunotherapy
Tumor immunology literature synthesis and hypothesis generation.
STACK
Works with your model and your compute.
OpenAI-compatible APIsAnthropic APIDesktop app and CLISSH / Modal workflowsWindows WebView2macOS WebKit
FAQ
What to know before trying it.
Is Wisp Science a new model?
No. It is an open-source desktop and CLI application that uses your own provider key with OpenAI-compatible, OpenAI Responses, or Anthropic APIs.
How is it different from a generic AI assistant?
It can execute local tools: Python, shell, file operations, MCP database calls, SKILL workflows, and persistent artifact tracking.
Does research data stay private?
Raw files, sessions, artifacts, and settings are local. Prompts and model responses still go through whichever LLM provider you configure.
Is it production-stable?
It is a beta/MVP vertical slice. Core agent, streaming, tools, Python, MCP, and UI run, while multi-agent loops and 3D structure viewing are still roadmap items.
Try Wisp Science from the latest GitHub release.
Use the release page for current installers, or build from source if you want to inspect and modify the Rust/Tauri stack.