feat: add Vertex AI Grok support with provider abstraction

.env.example

@@ -12,3 +12,38 @@ GROK_API_KEY=your_grok_api_key_here
 
 # Optional: Telegram bot token for /connect → Telegram (see README)
 # TELEGRAM_BOT_TOKEN=123456:ABC...
+
+# ----------------------------------------------------------------------
+# Vertex AI Grok backend (optional)
+#
+# When GROK_PROVIDER=vertex (or `--provider vertex`) is set, the CLI
+# routes Grok traffic through Google Cloud Vertex AI instead of the
+# native xAI API. Authenticate locally with:
+#
+#   gcloud auth application-default login
+#
+# and verify with:
+#
+#   gcloud auth application-default print-access-token
+# ----------------------------------------------------------------------
+
+# Select the Vertex backend. Default is "xai".
+# GROK_PROVIDER=vertex
+
+# Required for Vertex mode. The Google Cloud project that hosts the
+# Vertex AI API call. Falls back to GCP_PROJECT_ID if unset.
+# GROK_VERTEX_PROJECT_ID=my-gcp-project
+
+# Optional. Vertex region for the partner-model endpoint. Default is
+# "global", which routes to the global Vertex endpoint as documented
+# in the Grok-on-Vertex model cards. Use a region (e.g. us-central1,
+# europe-west1) only when required by org policy or quota.
+# GROK_VERTEX_LOCATION=global
+
+# Optional. Override the Vertex API host (default: https://aiplatform.googleapis.com).
+# Rarely needed; useful for proxies or staging endpoints.
+# GROK_VERTEX_BASE_URL=https://aiplatform.googleapis.com
+
+# Authentication is currently ADC only (`gcloud auth application-default login`).
+# Pre-minted OAuth tokens and service-account-bound API keys are tracked as a
+# follow-up; no env var to set here for v1.

.npmignore

@@ -1,6 +1,9 @@
 # Source files (only include built dist/)
 src/
 tsconfig.json
+vitest.config.ts
+biome.json
+test-vertex-integration.ts
 
 # Lock files from other package managers
 yarn.lock
@@ -50,6 +53,9 @@ tests/
 *.test.ts
 *.spec.js
 *.spec.ts
+dist/**/*.test.*
+dist/**/*.spec.*
+dist/grok-standalone*
 coverage/
 .nyc_output/
 
@@ -62,6 +68,15 @@ docs/
 .git/
 .gitignore
 
+# Local development, agent, and editor runtime state
+.cursor/
+.omx/
+.codex/
+.agents/
+.claude/
+.grok/
+.husky/
+
 # CI/CD files
 .github/
 .gitlab-ci.yml
@@ -77,4 +92,4 @@ appveyor.yml
 
 # Include only what's needed for the package
 # The dist/ folder should be included (not ignored)
-# package.json and README.md should be included
+# package.json and README.md should be included

CHANGELOG.md

@@ -8,9 +8,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- **Vertex AI Grok backend.** Select with `--provider vertex` or `GROK_PROVIDER=vertex`. Authenticates via Google Application Default Credentials (`gcloud auth application-default login`), uses the documented Vertex Grok partner-model endpoint, and exposes the four official SKUs (`grok-4.20-{reasoning,non-reasoning}`, `grok-4.1-fast-{reasoning,non-reasoning}`). xAI behavior is unchanged for existing users; the active provider is settable via env var, the new `--provider` CLI flag, or `provider` in `~/.grok/user-settings.json`.
+- New `GrokProviderAdapter` contract at `src/providers/`. The Agent, tools, media, and helpers now depend on a backend-agnostic interface with explicit capability flags. xAI-only features (Batch API, hosted web/X search, image and video generation, Telegram STT) return typed errors instead of opaque transport failures when used under a Vertex session.
 - Dedicated grep tool powered by npm ripgrep WASM (#263)
 - `/btw` command for side questions (#264)
 
+### Fixed
+- TUI: plan-mode questions no longer leak into the chat input.
+- TUI: async tool-result and plan-question updates now repaint without waiting for a keyboard event.
+- TUI: typing a status nudge ("continue", "status") during an active turn now triggers a repaint instead of enqueueing an accidental follow-up message.
+- npm package no longer ships local agent/editor runtime state, stale dist artifacts, or platform-specific standalone binaries; the installed `grok` entrypoint now retains its executable bit across `tsc` builds.
+- Local PR-tested installs no longer get clobbered by the auto-updater pulling the latest public release.
+
 ### Changed
 - Switched Telegram voice/audio transcription from whisper.cpp to Grok STT (`/v1/stt`); removed `whisper-cli`, `ffmpeg`, and model-download requirements (#266, #265)
 - Install script warns when auto-resolving to a pre-release version (#269)

README.md

@@ -241,6 +241,75 @@ Optional: `**GROK_BASE_URL**` (default `https://api.x.ai/v1`), `**GROK_MODEL**`,
 
 ---
 
+## Vertex AI Grok (Google Cloud backend)
+
+`grok-cli` can route Grok traffic through **Google Cloud Vertex AI** instead of the native xAI API. The four documented Vertex Grok SKUs are supported:
+
+- `grok-4.20-reasoning` (200K context — default chat model)
+- `grok-4.20-non-reasoning`
+- `grok-4.1-fast-reasoning` (128K context)
+- `grok-4.1-fast-non-reasoning` (128K context — default title/recap model)
+
+### Prerequisites
+
+1. Enable the Vertex AI API on your GCP project.
+2. Authenticate locally with **Application Default Credentials**:
+   ```bash
+   gcloud auth application-default login
+   gcloud auth application-default print-access-token   # verify
+   ```
+
+### Run
+
+```bash
+# CLI flag
+grok --provider vertex
+
+# Or via env var
+GROK_PROVIDER=vertex GROK_VERTEX_PROJECT_ID=my-gcp-project grok
+
+# Or save in ~/.grok/user-settings.json
+{
+  "provider": "vertex",
+  "vertex": {
+    "projectId": "my-gcp-project",
+    "location": "global"
+  }
+}
+```
+
+### What works on Vertex
+
+- Chat and streaming
+- Local function/tool calling (bash, file edits, LSP, etc.)
+- Structured outputs
+- Image inputs (vision)
+- Reasoning vs non-reasoning by SKU selection
+
+### What's xAI-only (returns a typed error on Vertex)
+
+- xAI `/responses` endpoint and the multi-agent SKUs
+- Hosted web search and X search (`search_web`, `search_x`)
+- Image generation (`grok-imagine-image`) and video generation (`grok-imagine-video`)
+- xAI Batch API (`--batch-api`)
+- Telegram audio transcription via Grok STT
+
+### Vertex environment variables
+
+| Var | Purpose | Default |
+|---|---|---|
+| `GROK_PROVIDER` | Active backend (`xai` or `vertex`) | `xai` |
+| `GROK_VERTEX_PROJECT_ID` | GCP project for Vertex requests | — |
+| `GCP_PROJECT_ID` | Fallback for `GROK_VERTEX_PROJECT_ID` | — |
+| `GROK_VERTEX_LOCATION` | Vertex region (e.g. `us-central1`) | `global` |
+| `GROK_VERTEX_BASE_URL` | Vertex API host override | `https://aiplatform.googleapis.com` |
+
+> Authentication is currently **ADC only** (`gcloud auth application-default login`). Pre-minted OAuth tokens and service-account-bound API keys are documented in the Google Cloud quickstart but require additional CLI plumbing — tracked as a follow-up.
+
+> **Disclaimer:** Vertex AI access to Grok is provided by Google Cloud as a partner integration. This project is independent of both xAI Corp. and Google. Pricing and availability are governed by your Google Cloud contract.
+
+---
+
 ## Telegram (remote control) — short version
 
 1. Create a bot with [@BotFather](https://t.me/BotFather), copy the token.

package.json

@@ -15,8 +15,10 @@
   },
   "scripts": {
     "dev": "bun run src/index.ts",
-    "build": "tsc",
+    "clean:dist": "bun -e \"import { rmSync } from 'node:fs'; rmSync('dist', { recursive: true, force: true })\"",
+    "build": "bun run clean:dist && tsc && bun -e \"import { chmodSync } from 'node:fs'; chmodSync('dist/index.js', 0o755)\"",
     "build:binary": "bun build --compile --outfile dist/grok-standalone ./src/index.ts",
+    "prepack": "bun run build",
     "start": "bun run dist/index.js",
     "typecheck": "tsc --noEmit",
     "test": "bunx vitest run",
@@ -56,6 +58,7 @@
     "commander": "^12.1.0",
     "diff": "^8.0.3",
     "dotenv": "^16.6.1",
+    "google-auth-library": "^10.6.2",
     "grammy": "^1.41.1",
     "react": "^19.2.4",
     "ripgrep": "^0.3.1",