Cloud quickstart

Install the client

Follow Install the agent, then check:

thespider-agent --version

If thespider-agent setup --help does not list --model and --agent, run thespider-agent update first.

Run setup in the repository

cd /path/to/your/repository
thespider-agent setup --model xai/grok-4.6 --agent grok --profile security

--profile bugs is the other option. --agent also takes claude, codex, opencode, or any other name (pair an unknown name with --command).

Pass --agent on purpose. If you omit it, setup auto-detects and prefers Grok when several CLIs are installed. On a real terminal you can omit --model and search models.dev interactively. Scripts and AI agents must pass it.

Setup opens a browser login, creates or selects the matching project, enables the model, and writes .thespider/config.toml. It does not spend model tokens.

Approve one model call

Read the planned-call count in the setup summary. Then:

thespider-agent run --once --max-slices 1

That creates or reuses a full campaign and claims at most one slice. If more slices remain, the client prints 1 of N completed and the exact command to continue. One result is not a finished audit of the whole repo.

Open the dashboard

Setup prints a project URL. A completed slice may find issues or may correctly find none. Open that URL, pick the profile, and look at findings.

Once a project has a campaign, its page opens with a next step card: the single best thing to do now (retest, triage, remediate, or resume an unfinished campaign), why, and a prompt or command to paste into your coding agent. The organization dashboard ranks your projects the same way and offers one prompt per project, so you can pick up where you left off without rebuilding the state in your head. Every prompt is read-only first: your agent checks status and asks before any model call or status change. A prompt only ever carries ids, counts, and model codes, never finding text.