Overview
This series explains Grok Build as a layered, stateful agent runtime rather than a chat window. It traces how client requests become session turns, typed tool calls, workspace effects, permission and sandbox decisions, observable results, parallel tasks, and durable context. It ends with conditional guidance for TUI, headless, and ACP use. The repository supports architectural claims, not conclusions about model quality, latency, cost, adoption, or comparative security.
Learning outcomes
- Grok Build is a stateful agent runtime behind the TUI, headless mode, and ACP, not three separate systems.
- Sessions coordinate turns; typed tools and WorkspaceOps mediate how model intent reaches host effects.
- Permission rules authorize requests, while the process sandbox constrains reach afterward, with platform-specific limits.
- Parallel tasks and durable context remain useful when ownership, persistence, freshness, and current verification stay visible.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
What is Grok Build?
Grok Build is a terminal-based AI coding agent and stateful runtime that reads code, edits files, runs commands, searches, and manages ongoing work. TUI, headless mode, and ACP are different client boundaries over that runtime.
How does Grok Build turn a prompt into work?
Grok Build routes a prompt through an agent and session, lets the model select a typed tool, and sends the call through WorkspaceOps to a host context. Structured results and notifications inform the client and the next model turn.
How does Grok Build separate permission from sandbox safety?
Grok Build uses hooks and permission rules to decide whether a requested operation may run, while its OS-level sandbox limits what the process and children can reach afterward. Workspace checks and tool validation add further boundaries.
What does Grok Build preserve across a restart?
Grok Build can restore session updates, chat state, plans, task metadata, and related persisted state, while derived chat may be rebuilt from updates. The current workspace and retrieved memory still require verification.
Which Grok Build interface fits interactive, automated, or integrated work?
Use TUI for interactive review, headless mode for scripts or CI, and ACP when another application owns the presentation and permission flow. The underlying runtime and its controls still need explicit configuration.