Wisp 命令行

Once you know the desktop project and conversation workflow, you may want to use Wisp in a system terminal or connect a task to scripts and logs. The standalone wisp-science CLI provides that entry point.

This tutorial covers CLI preparation, model environment variables, interactive mode, and one-shot tasks. Wisp uses the current directory as its workspace, so check where your terminal is before starting. For desktop SSH environments and interactive terminals, see Server Environment Setup.

API addresses, model IDs, and file paths are examples. Replace them with your actual configuration. The commands are not evidence of completed analyses.

Check that the standalone CLI is available.

If you have built or installed the standalone CLI and wisp-science is on PATH, run it from your project directory. Installing the desktop app does not necessarily put this command on the system PATH.

Configure a model for this terminal.

The CLI uses environment variables. It does not automatically turn desktop keyring settings into terminal environment variables. For a compatible API, replace the placeholders below with your own values.

macOS / Linux:

export WISP_PROVIDER="openai"
export WISP_API_URL="https://your-api.example.com"
export WISP_MODEL="your-model-id"
read -s WISP_API_KEY
export WISP_API_KEY
wisp-science

After read -s WISP_API_KEY, type the key and press Enter. Input is not echoed. The URL and model ID above are placeholders, not a working connection.

Windows PowerShell:

$env:WISP_PROVIDER = "openai"
$env:WISP_API_URL = "https://your-api.example.com"
$env:WISP_MODEL = "your-model-id"
$credential = Get-Credential -UserName "api" -Message "Enter the API key in the password field"
$env:WISP_API_KEY = $credential.GetNetworkCredential().Password
wisp-science

Choose openai, openai_responses, or anthropic for WISP_PROVIDER according to the service's protocol.

Continue a conversation in the project directory.

After running wisp-science, enter natural-language tasks in interactive mode. For example:

Inspect the current project's directory without changing it. List likely data files, analysis scripts, and result directories. Do not install dependencies or modify files yet.

Use /help for help, /new for a new conversation, /compact to compact context, and /quit to exit.

Run one task or emit structured events.

From the project directory:

wisp-science run "Read-only: list the top-level project files and identify likely data, script, and result directories"
wisp-science run --output jsonl "Read-only: check the column names and missing values in data/example.csv"

jsonl emits one structured event per line for logs or scripts. These commands still call a real model. A command starting successfully does not establish that the example path exists or that analysis will succeed.

When running from source, check the working directory.

Developers can run cargo run -p wisp-cli -- run "task" from the repository root. This uses the repository directory as the workspace by default. To analyze another project, build the CLI and run the executable from that project's directory. See Development for builds and additional parameters.

Troubleshoot the command, configuration, and directory.

Start in a small practice directory and ask for a read-only file listing. Once the model responds and the directory is correct, try data inspection or one-shot output.

See CLI Development and Model Configuration. This tutorial reflects the implementation when written; parameters and messages may vary between versions.