Grok Build reaches 1.0 as SpaceXAI hardens its terminal coding agent

Why it matters

Grok Build 1.0 gives Musk a stable distribution layer for Grok 4.5 as AI labs compete to own the coding agent developers run inside real repositories.

Elon Musk (@elonmusk) said in a post on X on August 7th that Grok Build V1.0 had shipped, moving SpaceXAI's terminal coding agent out of its 0.x beta after roughly 10 weeks.

The major-version label suggests maturity. The published changelog shows a release centered on reliability, permissions and interface repairs rather than a new model or headline coding capability. SpaceXAI fixed failure points involving large repositories, remote sessions, queued messages, authentication, background tasks and permission prompts.

That distinction matters because Grok Build already received its larger technical upgrades before 1.0. SpaceXAI made Grok 4.5 the default model in July, opened the agent harness under an Apache 2.0 license and added workflows that can distribute jobs across large groups of parallel agents. Version 1.0 packages that expansion behind a more stable release boundary.

A 1.0 release built around operational trust

The Grok Build 1.0 changelog lists more than two dozen changes. Permission prompts now display complete scripts, including long Bash bodies. Remote resume restores the conversation without changing local code unless a user passes --restore-code. Codebase restoration no longer hangs on large or shallow Git repositories, and Grok retries a wider range of server errors during outages.

SpaceXAI also addressed cases where queued prompts disappeared, background tasks restarted the model after cancellation, invalid API keys interfered with login and large session forks consumed excessive memory. Dashboard rows now summarize the agent's previous turn, while the model picker remains usable during plan review.

These fixes define the practical stakes of the 1.0 designation. Grok Build can read and edit files, execute shell commands, search the web and run long-lived tasks. A coding agent with that access needs predictable cancellation, visible scripts and recoverable state before teams can place it inside routine development work. SpaceXAI's permissions documentation separates approval modes from sandbox controls and allows administrators to define explicit allow and deny rules for tools and commands.

SpaceXAI expanded the harness before declaring 1.0

SpaceXAI introduced Grok Build in beta on May 25th as a terminal agent for SuperGrok and X Premium Plus subscribers. The initial product combined a plan-review interface, parallel subagents, headless operation for scripts and support for the Agent Client Protocol.

The current Grok Build product page says the agent is available to try free and runs on Grok 4.5. It can load project instructions, skills, plugins, hooks and Model Context Protocol servers, while its plan mode blocks edits until the user approves a proposed implementation.

SpaceXAI open-sourced the harness on July 15th. The GitHub repository exposes the Rust code for the terminal interface, agent runtime, tool dispatch, workspace management and extension system. The repository showed about 24,300 stars on August 7th, a measure of developer attention rather than active usage or paid adoption.

The open-source label also has a boundary. SpaceXAI says the public repository is periodically synchronized from an internal monorepo and does not accept external contributions. Developers can inspect, compile and modify the Apache-licensed code, including pointing it at local inference, while SpaceXAI retains control over the upstream product roadmap.

On July 23rd, SpaceXAI added workflows, which let Grok Build create orchestration scripts, send parts of a job to parallel agents and combine their results. Standard runs receive a budget of 128 agents, according to SpaceXAI, with as many as 1,024 available for larger jobs. The capability targets work such as repository-wide audits, issue triage and multi-stage code review.

SpaceXAI had already changed the product's model layer on July 16th, when it introduced Grok 4.5 and made the model Grok Build's default. The company positions Grok 4.5 for coding, agentic tasks and technical knowledge work. SpaceXAI's benchmark and efficiency claims remain company-reported; the 1.0 release adds no fresh performance measurements.

The contest is shifting to the agent shell

Grok Build enters 1.0 in a market where model providers increasingly control both the underlying model and the software developers use to direct it. Anthropic's Claude Code reads repositories, edits files and runs tests. OpenAI's Codex CLI provides local code modification and execution with configurable approval modes. Google released the open-source Gemini CLI for terminal-based coding and task execution.

SpaceXAI is competing through an extensible harness that can run its own model or a custom one. The Grok Build documentation allows developers to configure third-party model endpoints, use the agent headlessly in automation and embed it in other applications through ACP. Grok Build also reads Claude Code instruction files, skills, plugins and hooks, lowering the switching cost for developers who already maintain those configurations.

Version 1.0 does not introduce a new pricing plan, benchmark result or model. It gives Musk a stable product marker for the distribution layer around Grok 4.5. SpaceXAI's next test is whether the combination of open code, compatibility and parallel-agent workflows can convert developer curiosity into sustained use inside production repositories.