* feat: add CLI update checker
Check for newer grok-dev releases when the interactive CLI starts and surface them through a startup modal, home banner, and update commands.
* ci: increase PR commit scan curl timeout to 300s
Expose Shuru sandbox configuration beyond the simple on/off toggle.
Settings (allowNet, allowedHosts, ports, cpus, memory, diskSize, secrets,
checkpoint) flow from CLI flags / saved settings through Agent, sub-agents,
and background delegations into wrapCommandForShuru. The TUI /sandbox panel
is replaced with a unified settings modal using arrow left/right for toggles
and inline editing for text fields. Adds --allow-net, --allow-host, and
--port CLI flags and a README section on sandbox support.
* feat: add Shuru sandbox mode for agent shell execution
Adds a user-selectable sandbox mode that runs all agent-driven bash
commands inside a Shuru microVM. Includes persistent user/project
settings, CLI flags (--sandbox / --no-sandbox), a TUI picker via
/sandbox, runtime toggling for live agents, delegation propagation,
model prompt updates, and Phase 1 guardrails that block shell commands
which would silently lose workspace mutations inside the guest overlay.
* fix: use word boundary in git guardrail to catch compound commands
The git mutation guard used ^ (start-of-string anchor), so compound
expressions like `echo foo && git commit` bypassed it. Switch to \b
to match the pattern used by the other guardrails.
* fix: remove branch from git allowlist and stabilize bash tests
- Remove `branch` from the read-only git allowlist since git branch
can create/delete/rename branches, which are mutating operations.
- Export wrapCommandForShuru and getSandboxMutationBlockReason for
direct unit testing without child_process mocking.
- Rewrite bash.test.ts to test pure functions directly, eliminating
flaky vi.doMock/resetModules ESM interactions.
* feat: add scheduled headless runs with daemon and agent tools
Add schedule_create, schedule_list, schedule_remove, schedule_read_log,
schedule_daemon_status, schedule_daemon_start, and schedule_daemon_stop
agent tools. Schedules are persisted under ~/.grok/schedules/ and fired
by a lightweight `grok daemon` background process. Includes /schedule
TUI slash command, browser modal, README section, and tests.
* fix: resolve pre-existing test failures across the suite
Add vitest.config.ts to exclude dist/ from test runs. Fix whisper-cpp
and instructions tests to isolate from local filesystem state by setting
process.env.HOME to a temp dir. Fix bridge test lint errors by replacing
`any` with proper types.
* fix: address code review findings in schedule system
- Prevent path traversal in schedule IDs by stripping dots from allowed
chars and adding assertInsideSchedulesDir guard
- Fix day-of-week cron ranges including 7 (e.g. 5-7) by validating raw
values before normalization and handling wrap-around in matching
- Pass log file descriptor directly to child stdio instead of piping
through the parent process
- Isolate per-schedule errors in daemon tick so one broken schedule
does not block the rest
* ✨ Add built-in telegram-bridge command
- add a headless Telegram bridge entrypoint and reusable path resolver
- export the new bridge helpers and cover path defaults with focused tests
- document headless Telegram usage in the README and verify startup without the wrapper
* 🐛 Fix telegram bridge shutdown handling
- restore the CLI's default SIGTERM exit behavior for non-bridge commands
- let the headless telegram bridge own SIGTERM cleanup while it is running
- abort bridge startup and clear watchers/listeners when Telegram setup fails early
* ♻️ Share Telegram approval persistence
- move approved-user persistence into a shared Telegram settings helper
- reuse the helper from both the TUI pairing flow and the headless bridge
- keep the approval write path consistent while removing duplicated settings merge logic
* bump package.json version
---------
Co-authored-by: Ismail Pelaseyed <homanp@gmail.com>
* feat: add telegram_send_file tool for Telegram media attachments
Give the agent a real tool to upload files to Telegram chats so it can
confirm delivery before writing its final reply, instead of guessing
whether attachments are supported.
* feat: add Telegram voice/audio transcription via whisper.cpp
Local speech-to-text for Telegram voice notes and audio attachments
using whisper.cpp, with auto model download and ffmpeg conversion.
* feat: add Telegram voice/audio transcription via whisper.cpp
Telegram voice notes and audio attachments are now transcribed locally
using whisper.cpp before being sent to the agent. Includes auto-download
of Whisper models, ffmpeg fallback for non-WAV formats, configurable
settings under telegram.audioInput, and unit tests for the new STT layer.
* fix: harden whisper audio setup
Avoid Linux model download failures, remove unused STT dependencies, and keep local Cursor state changes isolated to the intended files.
* refactor: share Telegram audio handler
Reduce duplicate Telegram voice and audio handling so transcription flow changes stay in sync.
Unify Telegram turn handling with the interactive TUI so remote messages show live tool calls, tool results, and subagent activity instead of only final responses.
* fix: only send reasoningEffort for grok-3-mini
According to xAI documentation, only grok-3-mini supports the
reasoning_effort parameter. Other models (grok-3, grok-4, grok-4-fast-reasoning)
do not support this parameter and return a 'Bad request' error when it is sent.
Changes:
- Added supportsReasoningEffort field to ModelInfo interface
- Marked only grok-3-mini with supportsReasoningEffort: true
- Updated getSupportedReasoningEfforts() to check supportsReasoningEffort
instead of the generic reasoning flag
- This prevents sending reasoningEffort to unsupported models
Fixes the 410 error when using models like grok-code-fast-1 with
reasoning effort configured.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* test: update tests to reflect grok-3-mini-only reasoning effort support
Updated model tests to verify that only grok-3-mini supports
reasoningEffort parameter, and all other models (including reasoning
models like grok-4-0709, grok-code-fast-1) correctly return empty
supported efforts arrays.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* test: add integration tests for resolveModelRuntime
Verify that providerOptions.xai.reasoningEffort is only included
for models that support it (currently only grok-3-mini), and is
correctly omitted for all other models regardless of their
reasoning flag.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* fix: organize imports in client.test.ts for Biome linter
Biome requires imports to be sorted alphabetically. Moved @ai-sdk/xai
import before vitest import to satisfy the linter.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* fix: add comprehensive tests for reasoningEffort with grok-3-mini
- Fixed misleading test name that said 'includes' but asserted undefined
- Added new test suite 'with configured reasoning effort' that mocks
getReasoningEffortForModel to verify positive cases
- Tests now verify that grok-3-mini DOES receive providerOptions when
reasoning effort is configured (high or low)
- Tests verify that other models (grok-4-0709, grok-code-fast-1, etc.)
do NOT receive providerOptions even when effort is configured
- Increased test coverage from 90 to 100 tests
Addresses review feedback about missing positive test case.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* refactor: remove dead code branches in getSupportedReasoningEfforts
Removed unreachable multiAgent and responsesOnly branches since only
grok-3-mini currently has supportsReasoningEffort: true, and it has
neither of those flags.
Added comment explaining that only grok-3-mini supports reasoning_effort
per xAI documentation, and it supports 'low' and 'high' efforts.
If xAI adds reasoning_effort support to other models in the future,
the logic can be expanded as needed.
Addresses review feedback about dead code.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: add custom sub-agents with /agents TUI and reliable interrupt
- Custom sub-agents stored in ~/.grok/user-settings.json (name, model, instruction)
- Validation against known model IDs, reserved name protection
- Injected into system prompt and available via task tool in all modes
- /agents TUI modal for browsing, adding, editing, and removing sub-agents
- Custom agent name shown in task UI instead of generic "Sub-agent"
- Foreground bash/search tools now accept abort signals for real cancellation
- Esc interrupt aligned with queue-force path for consistent cancel behavior
* fix issues found in code review
The CLI was showing version "1.0.0" but package.json has "1.0.0-rc2".
Changes:
- Import package.json in src/index.ts
- Use packageJson.version in program.version()
- Ensures --version shows correct version matching package.json
Fixes incorrect version displayed to users.
Co-authored-by: Ren <ren@openclaw.ai>
* feat: Agent Skills discovery and prompt catalog
- Discover skills in ~/.agents/skills and project .agents/skills (project wins on name)
- Inject OpenCode-style <available_skills> XML + read_file instructions in system prompt
- /skills slash command lists installed skills in the TUI
- Document layout and usage in README; gitignore .agents for local skill installs
* chore: ignore skills-lock.json
* feat: add startup API key modal
Allow the interactive app to boot without a preconfigured API key and guide users into adding one in-app instead of exiting early. Persist explicit auth and model preferences so first-run setup and subsequent launches feel consistent.
* bump version
* format
* fix lint issues
* feat: add background agents and background bash processes (#161)
Add async background delegation system (delegate, delegation_read,
delegation_list) for read-only research that persists to disk and
notifies the session on completion. Extend bash tool with background
mode for long-running processes (dev servers, watchers) with companion
process_logs, process_stop, and process_list tools.
Cherry-picked from a10a337 — lost during merge of cursor/agent-new-architecture-b3d5.
* feat: add prompt queuing with follow-ups UI
Allow users to submit messages while the agent is processing. Queued
messages display in a follow-ups panel above the prompt and are
processed sequentially after the current response completes.
Cherry-picked from 299023b — lost during merge of cursor/agent-new-architecture-b3d5.
* chore: switch CI to Bun, remove unused files and dead code
- Update security workflow to use Bun instead of npm (fixes ERESOLVE peer dependency conflict)
- Delete unused files: hello.py, package-lock.json, .eslintrc.js
- Remove broken ESLint config and devDependencies (lint was non-functional)
- Remove unused exports: MODEL_GROUPS, Skill interface, formatDuration
* chore: update lockfile after removing ESLint dependencies
* fix: use ref for mode color to prevent stale closure in message borders
* feat: add Biome for linting and formatting, update CI workflow
- Add Biome as linter and formatter with project-appropriate config
- Add lint, format, format:fix, and lint:fix scripts to package.json
- Rename typecheck workflow to CI, switch from npm to Bun, add format
and lint steps alongside typecheck
- Auto-fix formatting and import ordering across all source files
* fix: improve cancellation flow and queued message editing
* fix: improve cancellation flow and prompt footer context display
* chore: update Bun version and refresh lockfile
* feat: complete architecture rewrite with Bun, OpenTUI, bash-only tools, and X Search API
Major rewrite of the Grok CLI agent:
- Runtime: Bun-first (removed Node.js/tsx fallbacks)
- UI: Replaced Ink with OpenTUI (@opentui/react) terminal framework
- Tools: Simplified to bash-only (removed text-editor, search, morph, todo, confirmation tools)
- Search: Integrated X Search API and Web Search via xAI Responses API
- Models: Updated to all latest Grok models (grok-4, grok-4.20-beta, grok-4-1-fast, grok-4-fast, grok-3-mini)
- Agent: Streamlined agent loop with async generator streaming
- Client: Dual API support (Chat Completions for tool calling, Responses API for search)
- Removed MCP integration, confirmation service, token counter, Morph editor
- Updated tsconfig for OpenTUI React JSX
- Clean 10-file architecture in src/
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* Fix HelpPanel rendering bug: replace <br/> with Text component
- Replaced <br/> tag with <text>{" "}</text> in HelpPanel component
- Ink requires all HTML-like elements to be wrapped in Text components
- Fixes React reconciler error: 'Component of type br must be created inside of a Text node'
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* fix: use ref for reliable streaming content accumulation in UI
- Replace fragile setState callback with contentAccRef for tracking streamed content
- Save intermediate assistant content when tool calls arrive
- Remove unused useTerminalDimensions import
- Fix HelpPanel <br/> tag (must be inside text node in OpenTUI)
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* fix: improve UI visibility with dark theme and text wrapping
- Add dark background (#0f172a) to main app, model picker, input
- Set explicit text/background colors on input for visibility
- Use wrapMode='word' on all text content for proper wrapping
- Use children instead of content prop for user messages
- Add Grok label to streaming content display
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* fix: remove wrapMode that caused garbled text rendering
wrapMode='word' was corrupting character display in OpenTUI text components.
Removed it from all message views and streaming display.
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: rewrite UI to match OpenCode's design language
- Use OpenCode's exact dark theme tokens (#0a0a0a bg, #141414 panel, #eeeeee text, #fab283 primary)
- Left-border accent (┃) on user messages with primary color, matching OpenCode's SplitBorder
- Panel backgrounds for user messages and tool blocks (backgroundPanel)
- Home screen with centered logo + prompt (like OpenCode's home.tsx)
- Session view with header, scrollable messages, prompt, footer
- Inline tool results (icon + label) for simple tools
- Block tool results (panel with output) for bash with >3 lines
- Footer with cwd on left, model on right (matching OpenCode's footer.tsx)
- Separate theme.ts with typed dark theme object
- Reasoning shown in subtle left-bordered panel (like OpenCode's ReasoningPart)
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: add Agent/Plan/Ask mode system with tab cycling and OpenCode home screen
- Add AgentMode type (agent/plan/ask) with per-mode system prompts
- Agent mode: full tool access, executes changes directly
- Plan mode: read-only tools, creates implementation plans without modifying files
- Ask mode: read-only tools, answers questions with context
- Tab key cycles between modes (visible in prompt box)
- OpenCode-style block-letter GROK logo on home screen (left=muted, right=bold)
- Prompt box shows mode label + model name + provider (like OpenCode's prompt)
- Shadow/bottom-edge effect below prompt (╹ + ▀ pattern from OpenCode)
- Keyboard hints bar: 'tab modes · ctrl+p commands' (matches OpenCode layout)
- Mode color changes left-border accent on prompt and user messages
- Refactored into HomeView and SessionView components
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: black and white theme, Grok brand logo, modal model picker
- Theme: pure black (#000000) and white (#ffffff) only, no color accents
- Logo: Grok brand symbol (stylized compass/G) next to bold 'Grok' text
- Model picker: floating modal dialog overlaying the app (like OpenCode's DialogSelect)
- Centered panel with dark background
- Title 'Select model' + esc hint
- Search placeholder
- Models grouped by category (Grok 4, Grok Code, Grok 3)
- Current model marked with ●
- Selected item highlighted with selectedBg
- Pricing shown for budget models
- Mode colors changed to white/gray (monochrome)
- Models grouped via MODEL_GROUPS in models.ts
- Added selectedBg and selected tokens to theme
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: differentiated UI, colored modes, clean logo, remove xAI labels
- Remove all 'xAI' labels from model picker and prompt box
- Mode colors: blue (#5c9cf5) for Agent, yellow (#e5c07b) for Plan, orange (#f5a742) for Ask
- Logo: clean '✦ G R O K' text with tagline (no broken Unicode art)
- Differentiated UI from OpenCode:
- Home: centered logo + tagline, input with thin single-line border colored by mode,
mode selector row showing all three modes with dots, model name on right
- Session: top bar with model + mode instead of bordered header panel,
thin separator line, user messages with colored ▸ prefix instead of left-border panels,
input with colored single-line border instead of thick ┃ accent
- Tool output in subtle background panels
- Model picker: bordered modal instead of borderless floating panel,
category headers in muted text, no 'xAI' provider label
- Simplified theme: removed unused color tokens, added textDim for ultra-subtle text
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* revert: restore ┃ left-border prompt box with mode/model line and shadow edge
Bring back the previous input style that was preferred:
- ┃ left-border accent colored by mode (blue/yellow/orange)
- Mode label + model name inside the prompt box below input
- ╹/▀ shadow edge below the prompt
- Hints row (tab modes · ctrl+p commands) right-aligned under prompt
- Used in both home screen and session view
- Keep everything else from current version (differentiated session layout,
▸ prefix messages, top bar, clean model picker, logo, mode colors)
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: replicate OpenCode conversation layout
Session view now matches OpenCode's session/index.tsx patterns:
- Session header: # Model Name in ┃ left-border panel with backgroundPanel
- User messages: ┃ left-border colored by mode, backgroundPanel fill, text inside
- Tool results (inline): paddingLeft=3, icon + muted text ($ cmd, ◈ Search)
- Tool results (block): ┃ left-border panel for bash output >3 lines, search results
- Assistant text: paddingLeft=3, marginTop=1, plain text
- Metadata line: ▣ Mode · model · duration after each assistant turn (like OpenCode's
AssistantMessage final block with ▣ icon)
- Reasoning: ┃ left-border panel with backgroundElement border, italic dim text
- Pending tools: ~ prefix with text color
- Track tool call count and elapsed time for metadata line
- InlineTool component for consistent tool result formatting
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: redesign model picker and slash menu modals
- Remove borders, add search filtering with keyboard input
- Flat model list without category sections
- Blue accent color for current model highlight
- Fit-to-content modals with max height cap and scroll
- scrollChildIntoView for arrow key navigation
- Trim logo height by removing decorative lines
- Add accent color to theme
Made-with: Cursor
* feat: migrate to Vercel AI SDK with xAI provider
- Replace custom OpenAI client with AI SDK streamText + @ai-sdk/xai
- Add standalone search_web and search_x tools via chat completions
- Fix follow-up messages by using isProcessingRef to avoid stale closures
- Fallback message history when result.response fails after stream errors
- Remove post-response status line, adjust input padding
Made-with: Cursor
* feat: add file tools, diff view, and markdown rendering
- Add read_file, write_file, edit_file tools with diff generation
- Inline diff viewer with line numbers, colored add/remove, and
unmodified-lines separators (OpenCode-style)
- Lightweight markdown renderer for assistant messages: headings,
bold, italic, inline code, code blocks, lists, blockquotes, links
- Update agent system prompts with terminal environment guidance
- Install diff package for unified diff computation
Made-with: Cursor
* feat: add plan mode (#156)
* feat: add plan mode with interactive questions UI
Add generate_plan tool, structured plan types, and an OpenCode-style
inline questions panel with tabbed navigation, number-key shortcuts,
custom text answers, and a review/confirm tab.
* fix: plan-to-agent handover, mode-scoped tools, and UI polish
- Carry plan context forward when switching plan→agent mode
- Restrict generate_plan tool to plan mode only
- Stamp modeColor on all message types for correct per-message colors
- Sanitize model-echoed role labels and raw JSON from content
- Truncate tool result labels to single lines
- Single-word tab labels for plan questions
- Always show edit/write file results with diff
- Clear agent history on mode switch
* feat: add Brin contributor trust check workflow and use OpenTUI native markdown (#157)
* feat: add sub-agent task delegation and UI (#159)
* feat: add background agents and background bash processes (#161)
Add async background delegation system (delegate, delegation_read,
delegation_list) for read-only research that persists to disk and
notifies the session on completion. Extend bash tool with background
mode for long-running processes (dev servers, watchers) with companion
process_logs, process_stop, and process_list tools.
Closes#146
* feat: add prompt queuing with follow-ups UI (#162)
Allow users to submit messages while the agent is processing. Queued
messages display in a follow-ups panel above the prompt and are
processed sequentially after the current response completes.
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Ismail Pelaseyed <homanp@users.noreply.github.com>
* feat: add plan mode with interactive questions UI
Add generate_plan tool, structured plan types, and an OpenCode-style
inline questions panel with tabbed navigation, number-key shortcuts,
custom text answers, and a review/confirm tab.
* fix: plan-to-agent handover, mode-scoped tools, and UI polish
- Carry plan context forward when switching plan→agent mode
- Restrict generate_plan tool to plan mode only
- Stamp modeColor on all message types for correct per-message colors
- Sanitize model-echoed role labels and raw JSON from content
- Truncate tool result labels to single lines
- Single-word tab labels for plan questions
- Always show edit/write file results with diff
- Clear agent history on mode switch
- Replace custom OpenAI client with AI SDK streamText + @ai-sdk/xai
- Add standalone search_web and search_x tools via chat completions
- Fix follow-up messages by using isProcessingRef to avoid stale closures
- Fallback message history when result.response fails after stream errors
- Remove post-response status line, adjust input padding
Made-with: Cursor
Ismail Pelaseyed