What it does
7 features
Grok Build · Terminal coding agent with open source code
Product and access
Grok Build is not Grok Bot. Grok Build runs on your computer in the terminal. Grok Bot, listed separately in the catalog, works according to its docs with named Bots on a persistent cloud computer in Cursor's cloud and through desktop and mobile apps. The Grok Bot page does not mention Grok Build. According to the Grok FAQ, Grok Build replaces the no longer supported Grok Studio.
Core features
7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
Fullscreen terminal
According to the README, the interface understands the codebase, edits files, runs commands, searches the web and manages long-running tasks.
Source 7Headless and ACP
`grok -p` runs without an interface in scripts and CI, optionally with JSON output. Through the Agent Client Protocol the agent can be embedded in other applications.
Source 10Plan mode and goals
According to the docs, in plan mode only the plan file can be edited until you approve. `/goal` works toward an objective across rounds until reviewers consider it complete.
Source 11Subagents
Subagents are on by default and run as separate sessions with their own context. They report a summary back to the parent session.
Source 12Recurring tasks
`/loop` and the scheduler start prompts at intervals as background subagents. Recurring tasks expire after seven days, and at most 50 can be active at once.
Source 13Permissions and sandbox
Modes and rules control what runs without asking. The optional sandbox uses Landlock on Linux and Seatbelt on macOS.
Source 14
Show all 7 features
Enterprise controls
Managed configuration, enforced rules through `requirements.toml`, OIDC sign-in, team pinning and Zero Data Retention at team level.
Source 15
Usable AI models
Open choice
Grok models from xAI are the default. According to the docs, Grok Build supports any custom model: you add the base URL and key in config.toml, and the guide shows Anthropic, OpenAI, Bedrock, Ollama, Together AI and local servers. In a company, a central policy can pin the list.
Model families named
- xAI (Grok)
- Anthropic (Claude)
- OpenAI (GPT)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
- Depends on
- Admin approval
The rating covers the generally available main access. Documented · Sources accessed on
Grok Build: Overview · Grok Build: Custom models · Grok Build: Authentication
Does it fit my task?
2 practical tasks · 5 requirements and limits
A good fit if …
- you want a terminal agent with plan mode, subagents and CI use
- you already use Grok models or a Grok subscription
- your company wants to enforce rules centrally through configuration
Less suitable if …
- you want an agent that keeps running in the cloud without your computer; for that the provider describes Grok Bot
- you want to contribute to the code yourself
- you need to confirm location and retention commitments for Europe from public pages without reviewing a contract
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Implement a change with a plan
- Your input
- A project folder and a change description in plan mode.
- Possible result
- Grok Build writes a plan and implements it after your approval.
- What to check
- Read the plan and check the diff. `/rewind` only resets the conversation, not the files.
Run a task in CI
- Your input
- `grok -p` with `XAI_API_KEY` and rules through `--allow` and `--deny`.
- Possible result
- The agent runs the task and writes the result to standard output.
- What to check
- Use `dontAsk` mode or deny rules and evaluate the exit code.
Have it fix a failing test in the project folder, approving the plan first.
Your possible prompt
Read the project rules. Investigate [error], propose a small change, and run the relevant checks. Show me the diff and any remaining uncertainties.
Adjust the task, data and permissions.
Requirements and limits
Requirement
Account or API key
On first start you sign in through the browser; in CI you use `XAI_API_KEY`. You add custom models in `~/.grok/config.toml`.
Source 16Requirement
Requirements for the source build
Rust, DotSlash and protoc. macOS and Linux are supported build hosts; Windows builds are best effort and currently not tested from this tree.
Source 7Limitation
Open code without open contributions
The code is under Apache-2.0 but is only synced periodically from an internal monorepo. External pull requests are excluded. The latest sync in the repository dates from 29 September 2026; the npm package was at 1.0.70 when fetched.
Source 17Limitation
Experimental parts
Memory counts as experimental and is off by default, and OpenTelemetry monitoring has alpha status. The readable pages name no beta notice for the product as a whole.
Source 18Not yet confirmed
Retention without Zero Data Retention
The enterprise page gives no periods for organizations without Zero Data Retention. The 30 days in the API FAQ apply there to API requests; the page does not say whether they also apply to subscription sign-in.
Source 15
Can I use it?
Availability still unclear
Europe & regional availability
The privacy policy and the Europe addendum on x.ai answered with HTTP 403 when fetched, so no statement from them is used here. The enterprise page, the API security FAQ and the Grok FAQ on docs.x.ai were read.
Reviewed scope: Terminal agent with model calls through provider services
Reviewed on:
Availability still unclear
Availability and data location
The enterprise page names neither a data location nor regions. The Grok FAQ contains no regional privacy rules. The binding legal texts were not readable.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: Source blocked when fetched; please check directly with the provider.
Availability still unclear
Zero Data Retention
With Zero Data Retention at team level, according to the page, the inference layer stores no prompts, code or responses. The local history stays in `~/.grok/`. This is not a regional commitment.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: Only for teams with the feature enabled.
Availability still unclear
API: no regional guarantee
According to the API FAQ, the default endpoint guarantees no processing region. Requests are stored encrypted for 30 days by default. The only regional endpoint named is in the United States; the page names no European one.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: The page concerns the API. It does not say whether the same applies to Grok Build's subscription sign-in.
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Local agent with provider sign-in or API key
Sources reviewed:
Show all 3 points
Memory and conversation continuity
5 of 6 criteria documented
Markdown files for global and per-repository scope; can be turned on, edited locally and searched.
Experimental cross-session memory, off by default
Sources accessed:
Documentation by criterion
StorageDocumented
Storage and scope
Experimental and off by default. Global and workspace notes live as Markdown locally under `~/.grok/`. Clones and worktrees of the same repository share one scope.
SelectionDocumented
Selection and timing
You tell Grok what to remember or use `/remember` with a review panel. At session end Grok saves a metadata summary, and `/flush` writes a model-generated one. New facts are captured as observations after each turn.
Edit & deleteDocumented
Inspect, edit and delete
`/memory` shows all files and can delete notes. You can edit the files directly. According to the docs, forgetting by request is best effort only; deleting in the file is the reliable way.
Access boundariesPartly documented
Agent and channel boundaries
The boundaries run between global and workspace. The page does not describe boundaries between individual agents or channels.
RetrievalDocumented
Retrieval
Grok reads topics at session start. A bounded index of both scopes enters the context once per session, and memory tools search further when needed.
New conversationsDocumented
Session reset and compaction
`/new` starts fresh. `/compact` and auto-compact shorten the history; `/flush` can save what matters beforehand. `/dream` consolidates observations into topics.
Cost and effort
6 plans · Prices as of 5 November 2026
Prices and included usage
Pricing checked
The X Subscription pricing page lists plans and prices but does not mention Grok Build. According to the FAQ, Grok Build draws from the weekly pool of the Grok products. x.ai/pricing was blocked when fetched.
| Plan | Price | Included |
|---|---|---|
| The X Subscription: Free | $0, limited usage | The page does not say whether Grok Build is included. Price source |
| The X Subscription: Plus | $8 per month or $80 per year | Base usage. Price source |
| The X Subscription: Premium | $30 per month or $300 per year | 4x usage. Price source |
| The X Subscription: Super | $100 per month or $1000 per year | 15x usage. Price source |
| The X Subscription: Ultra | $200 per month or $2000 per year | 50x usage. Price source |
| grok-4.7 (API) | $2 input and $6 output per 1M tokens below 200,000 prompt tokens; $4 and $12 above | Public API price for access through an API key. Price source |
- According to the API pricing page, Grok 4.7 Fast is only available in Cursor and Grok Build and is billed through the plan there.
- Third-party pricing was not used as evidence.
Setup
Install the prebuilt binary with the install script, the npm package `@xai-official/grok` or WinGet. The first start opens the browser for sign-in; without a browser you set `XAI_API_KEY`. The source build needs Rust, DotSlash and protoc.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Local process; model calls through provider services or custom models
Sources reviewed:
Editorial assessment
Grok Build describes sessions, rules and enterprise controls at length in its guide. You have to turn on the sandbox and memory yourself. The state of the public repository can lag behind the shipped program, because it is only synced periodically.
Show all 6 points
Documented
Files and working state
Grok saves each session automatically under `~/.grok/sessions/`, grouped by working directory. This includes history, tool calls, the plan and rewind points. `/delete` removes a session permanently.
EvidenceDocumented
Sign-ins
Tokens live in `~/.grok/auth.json` with owner-only permissions. The docs advise full-disk encryption. In CI `XAI_API_KEY` applies, and an active session takes precedence.
EvidenceDocumented with conditions
Resuming after failures
`grok --resume` picks sessions back up, and `/fork` branches them. `/rewind` only resets the conversation; files stay unchanged. The reviewed page does not describe automatic retries after failures.
Evidence
Evidence and open questions
25 sources · 4 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Europe: Availability still unclear
Requirements and limits: not yet confirmed
Sources and evidence
Read the repository with its user guide through the GitHub API, docs.x.ai and the npm registry. x.ai/cli, x.ai/build, x.ai/open-source, x.ai/pricing, the privacy policy and the Europe addendum answered with HTTP 403 and are not used as evidence. Feature claims come from the provider; example tasks are editorial. Our own hands-on tests are pending.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
The repository under xai-org names SpaceXAI as the developer, and parts of the docs on docs.x.ai write SpaceXSI. The company also offers Grok, Grok Bot and the Grok API.
- Operator
- SpaceXAI / xAI
- Organization
- Provider
- Website
- x.ai
Changes to this profile 1
Terminal coding agent from SpaceXAI / xAI: it reads the codebase, edits files, runs commands and works interactively, headless or through ACP.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.