Getting Started

Getting Started

Install

Grab a release from the Releases page. The installer bundles skills/ and mcp-servers/, so a release is self-contained — you do not need the source tree to get the 42 skills and the bioinformatics databases.

Windows needs the WebView2 Runtime. Windows 10/11 usually has the Evergreen Runtime and the installer acquires it when missing; an outdated or damaged Runtime can still keep the main window from appearing. See the README's Windows installation and startup troubleshooting section.

macOS uses the system WebKit — nothing extra.

First run

1. Add a model. Settings → Models. Wisp speaks three provider protocols:

   Provider setting	Talks to
   openai	Any OpenAI-compatible Chat Completions endpoint — DeepSeek, GLM, local servers, OpenAI itself
   openai_responses	OpenAI's Responses API
   anthropic	Anthropic's API

   Your key goes into the OS keyring, not the project database.

2. Create a project. A project is a directory plus its own conversations, Python/R environment, artifacts and research graph. Point it at a real working directory — the agent reads and writes there.

3. Ask for something. No setup ritual. Try:

   • "Load data/counts.tsv and show me the distribution of library sizes"
   • "Find recent papers on FX-cell and tell me which ones actually replicate"
   • "Fold this sequence and tell me if the design is self-consistent"

Set up the local runtime

Ask: "set up my local environment" — the local-env-setup skill handles it. It bootstraps:

• uv / Python — provisions a venv at .wisp/python/.venv on first use
• Node + scimaster-cli — required by the nine bear-* literature skills
• pixi — for multi-environment bioinformatics work

R is opt-in and manual. You need Rscript on PATH and the jsonlite package installed yourself. Wisp never installs R packages automatically.

Building from source

Only needed if you are developing Wisp.

rustup target add wasm32-unknown-unknown
cargo install --locked trunk
cargo install tauri-cli --version "^2"

Plus Rust stable 1.88+, uv, and on macOS xcode-select --install.

Headless CLI

Same agent loop, same skills, same Python/R REPLs, no window:

export WISP_API_KEY="<your provider key>"
export WISP_PROVIDER="openai"
export WISP_MODEL="deepseek-v4-pro"
cargo run -p wisp-cli

Where things are stored

What	Where
Projects, conversations, artifacts, settings	SQLite on your disk
API keys, SSH passwords	OS keyring — never the database
Skills	skills/, bundled into the installer
MCP servers	mcp-servers/, bundled into the installer
Python venv	.wisp/python/.venv in the project

Nothing leaves your machine except calls to the model provider you configured and the databases a query explicitly targets.

Next

• Built-in Tools — what the agent can do out of the box
• Skills — the 42 bundled workflows
• Execution Contexts — adding a GPU server
• Interface Guide — finding your way around