Grok Build v1.0.4
Grok Build v1.0.4 was released on .
Summary
Grok Build 1.0.4 makes it easier to manage your work with smarter task expiry notices and gives you more control over web searches through domain filtering. We've also improved the composer to preserve your keystrokes and prevent accidental submissions while you're still typin...
Release Context
- This was the 117th Grok Build release of 2026 tracked by Havoptic.
- It shipped 1 day after Grok Build v1.0.3 (August 12, 2026).
Shipped around the same time
Within 3 days of this release, other AI coding tools also shipped:
- Claude Code v2.1.233 (August 14, 2026)
- GitHub Copilot CLI v1.0.80 (August 14, 2026)
- Antigravity CLI v1.1.13 (August 14, 2026)
- Claude Code v2.1.232 (August 13, 2026)
- Claude Code v2.1.231 (August 13, 2026)
Compare Grok Build: vs Claude Code · vs OpenAI Codex CLI · vs Cursor · vs Gemini CLI · vs Kiro CLI · vs GitHub Copilot CLI · vs Devin Desktop · vs Antigravity CLI
Release Notes
Grok Build 1.0.4
- Recurring /loop tasks show a one-line expiry notice when they auto-expire after 7 days
- Web search can be restricted to allowed or excluded domains via [toolset.web_search] in config.toml
- Session search index can be disabled via GROK_SESSION_SEARCH or [features] session_search for hosts sharing $GROK_HOME
- /session-info supports drag-to-select and copy, plus c and y shortcuts
- New Follow-up Behavior setting in /settings allows queued messages to steer immediately instead of waiting for the turn to finish
- Tool commands and MCP servers receive a GROK_SESSION_ID environment variable for the current session
- Relative markdown links open existing files in the working directory when no matching generated media is found
- Queued messages no longer auto-submit while you are still editing them in the composer
- Sessions poisoned by rejected images heal permanently so future turns succeed without retrying the bad image
- Auto permission mode honors explicit always-allow grants and narrow allow rules from settings
- Subagent lifecycle events stay correct even when delivered out of order
- Keystrokes typed while Grok is starting, or while a plan is generating, are preserved in the composer
- Background tasks killed from the UI wake the model when needed instead of staying parked
- [stop] / Ctrl+C inside a subagent overlay or drill-in stops the focused child instead of the root session
- Hook failures show the first line of stderr instead of only the exit code
- Pasting text or dragging images while scrollback is focused focuses the composer and pastes there
- Dashboard shortcut works with Ctrl+4 in terminals without the Kitty keyboard protocol
- Pasting image-only screenshots works on Linux without a clipboard error
- Editing a queued prompt to a slash command like /btw runs the command instead of sending text to the model
- grok du and worktree commands work on Windows when only USERPROFILE is set
- Permission-mode changes on the welcome screen apply to the newly created session
- Finished subagent transcripts are evicted from memory and rebuilt from disk when reopened
Other Recent Grok Build Releases
- v1.0.13 (August 28, 2026)
- v1.0.12 (August 27, 2026)
- v1.0.11 (August 26, 2026)
- v1.0.10 (August 24, 2026)
- v1.0.9 (August 24, 2026)
All Grok Build releases · All AI coding tool releases on Havoptic