Docs

Grok Build

How LinkCode runs Grok Build through the grok CLI in headless mode, and what that mode costs you — no approvals, no tool cards, no resume.

Grok Build is the least integrated of the five supported agents. LinkCode drives the grok CLI in its headless mode, which trades interactivity for simplicity: it can't stop to ask you anything, and it doesn't report the individual tool calls it makes.

Set it up

Grok Build is the one agent LinkCode can't download for you. Install the grok CLI yourself, and make sure it lands somewhere LinkCode looks — the host's PATH, ~/.local/bin/grok, or ~/.grok/bin/grok. If it isn't found, threads fail at start with a message naming the same locations.

Sign in

LinkCode has no sign-in flow of its own for Grok Build — it uses whatever credentials the grok CLI already has. You can also connect an xAI account under Settings → Providers, which LinkCode passes to the CLI for that thread.

Approval

Grok Build never asks

Every tool runs without an approval prompt, and this cannot be switched — headless mode has no way to wait for an answer, so the policy menu shows a single non-switchable Bypass permissions entry. Don't point Grok Build at a checkout you wouldn't hand over.

Model and reasoning effort

One model, Grok 4.5, and three reasoning-effort levels: Low, Medium, and High (High is the default). Both apply from the next turn.

What you won't see

Because headless mode streams only text and reasoning, a Grok Build turn shows the agent's replies but not the individual reads, edits, or commands behind them. File edits render as plain text rather than diff cards, there are no slash commands, no $ shell passthrough, and no image attachments, and a stopped thread cannot be resumed — start a new one instead.

On this page