Wisp Science is an open-source, local-first desktop AI research workbench that gives scientists and research teams an agentic workspace for literature search, Python and R execution, scientific database queries, and manuscript preparation, all running against their own machines.

What it is

Wisp Science — WISP, short for Workspace for Intelligent Scientific Practice — is a Rust desktop application distributed under AGPL-3.0-only, aside from directories that note otherwise. It is local-first by design: project data, conversations, and credentials stay on the user's machines, and the user brings their own model, whether an OpenAI-compatible endpoint, an Anthropic model, or an external agent such as Codex or Claude Code driven over ACP. It sits in the AI & Machine Learning / AI Interaction & Interfaces category of this registry.

The problem it addresses is fragmentation in computational research. Literature searching, notebook execution, remote server access, artifact handling, and manuscript drafting normally live in separate tools, and the trail connecting them — which figure came from which run, which decision produced which draft — is lost. Wisp Science keeps that trail inside one project, with persistent Python and R kernels, registered compute hosts, reusable Skills, and a durable conversation history, so a researcher can return after a restart and find the full record intact.

Key capabilities

• An agent that reads and writes project files, runs shell commands, and loads reusable Skills described in SKILL.md, structured so that skills do not flood the prompt.
• Persistent Python and R kernels that retain variables across cells and conversational turns, with a separate isolated kernel per conversation so parallel sessions never share state.
• Compute registration for local, WSL, and SSH hosts, including hardware probing and long-running Runs submitted with live logs.
• Bundled MCP servers connecting to PubMed, GEO, and roughly 80 other scientific databases.
• A Publication Workspace that freezes manuscript revisions into verifiable Evidence Capsules, plus isolated exploration branches for trying a direction without touching the mainline.
• Offline previews for notebooks, PDFs, Office files, and images, alongside artifact attachment with @, saved-session search with #, and skill application with /.
• Secrets stored in the OS keyring rather than SQLite, with encrypted manual sync and project transfer that never run in the background.

Who uses it and how

• Computational biologists and bioinformatics researchers, who can open a bundled demo requiring no API key and follow a full RNA-seq trajectory before configuring anything.
• Groups working from laptop to remote server, registering local, WSL, and SSH hosts once and submitting long Runs to them while keeping logs visible.
• Labs that need an auditable record of how a manuscript revision was produced, using Evidence Capsules to freeze and verify it.
• Researchers under data-governance constraints, for whom local storage, OS keyring credentials, and manual rather than background sync are the deciding factors.
• Users running several investigations at once, relying on per-conversation kernel isolation so parallel sessions cannot contaminate each other's state.

Getting started

Open the download page at wispscience.com or the GitHub Releases page, choose the operating system and processor, and install the signed package — MSI or NSIS on Windows, a signed and notarized .dmg for Apple Silicon and Intel on macOS, or .deb and AppImage for x86_64 and aarch64 on Linux. Then add a model under Settings → Models and create a project; building from source and the CLI are covered in the development documentation.

How it compares

The facts provided name no paid products that Wisp Science replaces, so its closest named neighbours are the general coding agents Codex and Claude Code, which it does not compete with but can drive over ACP as interchangeable back ends. Beyond those two, this registry lists no directly comparable scientific workbench.

When to use it — and when not to

There is no server-side stack to operate: Wisp Science is a desktop application with signed installers, so a self-hoster does not run a database, object store, or SMTP service, though anyone building from source must follow the development documentation. Teams that require silent background synchronisation, or that intend to embed the code in a proprietary product, should look elsewhere, because sync is deliberately manual and the licence is AGPL-3.0-only with per-directory exceptions that make the exact scope worth checking. The project is also modest in adoption at roughly 1,150 stars, so users should expect a smaller community than the mainstream coding agents.

@software{xu2026wisp,
  author    = {Xu, Zhou-Geng},
  title     = {Wisp Science: a local-first AI research workbench},
  version   = {v1.5.0},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22009273},
  url       = {https://doi.org/10.5281/zenodo.22009273}
}