Grok Build Powerhouse
GUI for Grok Build CLI (incl. Grok 4.5) — right in VS Code / Cursor.
Not affiliated with or endorsed by xAI. Grok, Grok Build, and xAI are trademarks of xAI; this project uses those names only to describe what it's compatible with.
Drop open files in as @-context, run multiple sessions at once, keep resumable chat history, generate images & video inline, dictate by voice, render Mermaid and LaTeX, and drive agent loops with verify. If you'd rather stay in the editor than a terminal, this brings Grok Build's agent into your sidebar — plus a Neural graph layer the CLI doesn't have.
Why use this?
If you live in your editor, Grok Build Powerhouse puts Grok Build next to your code:
- Native Diff Theater — every proposed edit can open VS Code's diff editor
- Modes — Agent / Plan / Auto with a real plan write-gate
- Parallel sessions with status dots; resume, rename, search
- Queue or steer while Grok works
- Inline
/imagine media, paste/attach images
- Voice dictation (ffmpeg + xAI STT; say “grok send”)
- Math & Mermaid in the stream
- Unique: Neural graph · power bonds · agent-loop verify
The CLI does the heavy lifting; this is the GUI when you'd rather not live in a terminal.
Features
Permission cards with diff preview
When Grok wants to write a file or run a command, Agent mode can raise a permission card. Open the native diff, then Allow once / always / Deny.
Modes — Agent, Plan & Auto accept
Toolbar: Agent (ask), Plan (block writes until you approve), Auto (YOLO). Plan gate is enforced in the extension host.
Image & video generation
/imagine (and related tools) surface paths inline in the chat via webview URIs.
Paste or attach images
Grok sees the pixels (ACP image content blocks), not just a path.
Voice control
Microphone in the composer → live transcription. Setup: docs/voice-setup.md. Requires ffmpeg + signed-in CLI (or STT key).
File chips & @ context
Open files, explorer attach, selection range chips, @ popover.
Session history
Parallel live sessions with status dots; past sessions from ~/.grok/sessions; rename & search.
Queue or steer
Type while Grok works — messages queue (or steer mid-turn when the CLI supports it).
Subagents
Delegated tasks render as cards; open detail / sibling transcript when available.
Tool calls
Every read, edit & command inline; expand for details (including shell IN/OUT when present).
Math & LaTeX
\(…\), \[…\], $$…$$ via vendored MathJax (offline). Bare $…$ is not a delimiter (avoids mangling “it costs $5”).
Mermaid diagrams
Fenced ```mermaid blocks render offline via a vendored Mermaid build.
Model picker & reasoning effort
Bottom toolbar / commands: switch model and --reasoning-effort for new sessions.
Neural graph (unique)
Live synapse map of project bonds, power hubs, touched files, and learn directions — Hebbian memory over your session activity.
Agent loop
implement → run your verify command → retry up to N rounds (Ctrl+Shift+L).
Requirements
- VS Code 1.90+ (or Cursor / compatible). Secondary Side Bar is best on 1.106+.
- Grok Build CLI on macOS, Linux, or Windows (native Windows — no WSL required).
- Login: SuperGrok / X Premium+ (
grok login) or xAI API key. Free chat tier does not include the CLI agent.
- Voice (optional):
ffmpeg — docs/voice-setup.md.
Install
- Package & install (from this folder or monorepo):
cd vscode-extension
npm install
npm run package
code --install-extension .\grok-build-powerhouse.vsix --force
Or: npm run install:powerhouse from the monorepo root.
Open Grok — Ctrl+; (Mac: Cmd+;). The sidebar walks setup when the CLI is missing.
Sign in — grok login or API key. One click steps via Setup Checklist.
Full notes: docs/INSTALL.md.
Quick start
- Open Grok Build Powerhouse (
Ctrl+;).
- Type a prompt and press Enter — Grok streams; activity shows on the composer strip.
- Turn on Show thinking traces with
grokPowerhouse.showThinking if you want full reasoning blocks.
- Approve tools when in Agent mode; use Plan for gated planning.
- Resume from the clock / history panel anytime.
Configuration
All settings under grokPowerhouse.* (VS Code Settings → search “Grok Build Powerhouse”).
| Setting |
Default |
Meaning |
cliPath |
auto |
Path to grok |
defaultModel |
auto |
Model id |
defaultEffort |
"" |
--reasoning-effort |
defaultMode |
auto |
agent / plan / auto |
steerByDefault |
false |
Steer vs queue while busy |
showThinking |
false |
Inline reasoning traces |
voiceApiKey |
"" |
STT override (else login token) |
ffmpegPath |
"" |
Mic capture binary |
voiceStreaming |
true |
Live STT vs batch |
voiceSendPhrase |
grok send |
Hands-free submit phrase |
neuralGraph |
true |
Live neural rail |
diffTheater / inlineDiffs |
true |
Diff recording / stream cards |
Commands: docs/COMMANDS.md.
How it works
The extension speaks JSON-RPC over grok agent stdio and renders results. Grok owns sessions, memory, MCP, models, and tool execution. Details: docs/architecture.md.
Complementary stack
| Piece |
Role |
| Grok Build CLI |
Required runtime (xai-org/grok-build) |
| Grok Build Powerhouse (this) |
Human GUI + neural + agent loops |
| grok-mcp (monorepo) |
MCP tools for Claude Code / Copilot orchestration |
| phuryn extension (optional) |
Another MIT community GUI — can run side by side |
Known limits
- Diff preview compares proposed old vs new text; write lands after approval.
- Worktree UI not implemented (same as the community GUI reference).
- Remote device link not implemented (local IDE focus).
- Voice needs ffmpeg + network to xAI STT.
- Fork requires a CLI that supports
_x.ai/session/fork.
Privacy
No message content / code / paths are sent to extension analytics. Turns go through the CLI to xAI as usual. Voice deliberately sends audio to xAI STT — docs/privacy.md.
License & attribution
MIT. Voice and MathJax integration patterns are adapted from phuryn/grok-build-vscode by Paweł Huryn (MIT). See docs/attribution.md.
