Run Claude Opus 4.7, Gemini 3, and Codex Inside Google Antigravity — Without an API Key.

# Fresh install (any OS with Node.js 18+)
npm install -g @anthropic-ai/claude-code

# If already installed, upgrade in place
claude update

# Verify version (must be 2.1.111 or higher)
claude --version

# Print existing values at all three Windows scopes
Get-Content $PROFILE
$env:ANTHROPIC_API_KEY
$env:ANTHROPIC_BASE_URL
[Environment]::GetEnvironmentVariable("ANTHROPIC_API_KEY", "User")
[Environment]::GetEnvironmentVariable("ANTHROPIC_API_KEY", "Machine")

# Note any values found. DO NOT clear them.
# The launcher in step 3 will null them inside its own scope only.

# Print existing values from your shell profile
grep -E "ANTHROPIC|CLAUDE_CONFIG" ~/.zshrc ~/.bashrc ~/.profile 2>/dev/null
echo $ANTHROPIC_API_KEY
echo $ANTHROPIC_BASE_URL
echo $CLAUDE_CONFIG_DIR

# Note what's set. The launcher will isolate, not delete.

# Create the isolated profile dir
New-Item -ItemType Directory -Path "$env:USERPROFILE\.claude-max" -Force

# Append launcher function to your $PROFILE
Add-Content $PROFILE @"
function claude-max {
    `$env:CLAUDE_CONFIG_DIR = `"`$env:USERPROFILE\.claude-max`"
    `$env:ANTHROPIC_BASE_URL = `$null
    `$env:ANTHROPIC_API_KEY = `$null
    `$env:ANTHROPIC_AUTH_TOKEN = `$null
    claude @args
}
"@

# Reload profile in current shell
. $PROFILE

# Verify launcher exists
Get-Command claude-max

# Create the isolated profile dir
mkdir -p ~/.claude-max

# Append launcher function to ~/.zshrc (or ~/.bashrc)
cat >> ~/.zshrc << 'EOF'

# DDS Vibe Academy — Claude Max isolated launcher
claude-max() {
  CLAUDE_CONFIG_DIR="$HOME/.claude-max" \
  ANTHROPIC_BASE_URL="" \
  ANTHROPIC_API_KEY="" \
  ANTHROPIC_AUTH_TOKEN="" \
  command claude "$@"
}
EOF

# Reload
source ~/.zshrc

# Verify
type claude-max

# Launch isolated Max profile
claude-max

# Inside the session, log in
/login

# Browser opens. Sign in with claude.ai Max account.
# Decline API credit offers if prompted.
# Browser returns "Login complete" — close the tab.

# Inside claude-max session
/status
# Expected: Login method: Claude Max account
# Expected: Model: Default Opus 4.7 with 1M context
# Must NOT show: API credits, Console, Pay-as-you-go

/model
# Picker opens. Select claude-opus-4-7 if not default.

# Stress test the model identity itself
Reply with the exact string OPUS47_ACTIVE and nothing else.
# Expected response: OPUS47_ACTIVE

# In Antigravity terminal, project root open
claude-max

# First real prompt — proves filesystem access + Opus 4.7
List the files in this directory grouped by type.
Then tell me which model you are and whether you have 1M context available.

# Requires Node.js 18+ (20 recommended)
node --version

# Install globally (latest stable v0.32.x as of March 2026)
npm install -g @google/gemini-cli

# Verify install
gemini --version

# Run from any directory
gemini

# Step 1: pick a theme
# Step 2: select "Personal Google Account" auth
# Browser opens, sign in, accept terms
# CLI returns to terminal with prompt >

# Quick test
> What model are you?

# /exit to leave session
/exit

# Inside gemini session
/help            # list all commands
/stats           # tokens used this session (API key users only)
/mcp             # show connected MCP servers
/clear           # reset conversation history
@filename        # include file contents in next prompt
/exit            # quit

# Headless scripting (one-shot, no follow-ups)
gemini -p "What is the gcloud command to deploy to Cloud Run?"

# Structured JSON output
gemini -p "Explain this codebase architecture" --output-format json

# Requires Node.js 18+
npm install -g @openai/codex

# Verify install
codex --version

# Show available subcommands
codex --help

# Browser-based OAuth (default)
codex login

# Click "Sign in with ChatGPT"
# Select your API organization in the OpenAI dashboard
# Browser auto-creates API key, returns to CLI

# Confirm
codex login status
# Exit code 0 means logged in

# Alternative: device-code flow (no browser)
codex login --device-code

# Alternative: manual API key (for CI/CD only)
printenv OPENAI_API_KEY | codex login --with-api-key

# Pin model (gpt-5.3-codex is the dedicated coding variant as of Q1 2026)
model = "gpt-5.3-codex"

# Approval mode: "on-request" prompts for risky commands
# Alternatives: "untrusted", "auto", "never" (read full docs)
approval_policy = "on-request"

# Sandbox: workspace-write blocks network + outside-workspace writes
# Alternatives: "read-only", "workspace-write", "danger-full-access"
sandbox_mode = "workspace-write"

# Enable network inside the workspace sandbox if your tasks need it
[sandbox_workspace_write]
network_access = false

# Where credentials are stored
cli_auth_credentials_store = "keyring"  # or "file" or "auto"

# Interactive TUI (default mode)
codex

# Headless one-shot
codex exec --full-auto "Run unit tests and fix failures"

# Read-only audit (no writes, no commands)
codex -s read-only "Audit this codebase for security issues"

# Override sandbox network on demand (e.g. for npm install)
codex -c 'sandbox_workspace_write.network_access=true' "Install dependencies"

# CI/automation only (skip approvals + sandbox)
codex --dangerously-bypass-approvals-and-sandbox "Deploy"

# In ~/.codex/config.toml — add this block
[mcp_servers.shopify-dev-mcp]
command = "shopify-dev-mcp"
args = []

# Or via CLI
codex mcp add shopify-dev-mcp shopify-dev-mcp

# List configured servers
codex mcp list

# Add an HTTP-based MCP with bearer auth
codex mcp add my-server https://example.com/mcp \
  --remote-auth-token-env MY_SERVER_TOKEN

# macOS
brew install ollama
brew services start ollama

# Linux (and WSL)
curl -fsSL https://ollama.ai/install.sh | sh

# Windows: download installer from ollama.com
# After install, Ollama runs as a service on port 11434

# Pull a chat model (14B fits comfortably in 12GB VRAM)
ollama pull qwen2.5-coder:14b

# Pull a fast autocomplete model (1.5B)
ollama pull qwen2.5-coder:1.5b

# Pull embeddings for codebase context
ollama pull nomic-embed-text

# Verify endpoint is live
curl http://localhost:11434/api/version

# In Antigravity:
# 1. Press Ctrl+Shift+X (Cmd+Shift+X on Mac) to open Extensions
# 2. Search "Cline" — install the saoudrizwan publisher version
# OR
# 2. Search "Continue" — install the Continue.dev publisher version
# 3. Reload Antigravity if prompted
# 4. Click the new extension icon in the activity bar

name: DDS Local Sovereign Config
version: 1.0.0
schema: v1

models:
  # Chat + edit + apply (heavy lifting model)
  - name: Local Chat — Qwen 2.5 Coder 14B
    provider: ollama
    model: qwen2.5-coder:14b
    apiBase: http://localhost:11434
    roles:
      - chat
      - edit
      - apply
    defaultCompletionOptions:
      temperature: 0.3
      contextLength: 32768

  # Tab autocomplete (fast small model)
  - name: Local Autocomplete — Qwen 1.5B
    provider: ollama
    model: qwen2.5-coder:1.5b
    apiBase: http://localhost:11434
    roles:
      - autocomplete
    autocompleteOptions:
      debounceDelay: 250
      maxPromptTokens: 1024
      multilineCompletions: auto

  # Embeddings for @codebase
  - name: Local Embeddings
    provider: ollama
    model: nomic-embed-text
    apiBase: http://localhost:11434
    roles:
      - embed

context:
  - provider: code
  - provider: docs
  - provider: diff
  - provider: terminal
  - provider: folder
  - provider: codebase

rules:
  - Never reproduce DDS certifications inaccurately. The five are GOTS, GRS, OCS, PETA-Approved Vegan, Fair Trade.
  - Always verify Shopify schema range steps satisfy (max-min)/step >= 3 and (default-min) % step == 0.
  - Output full files, not snippets, unless explicitly asked otherwise.

# Create a Modelfile with extended context
cat > qwen-cline-modelfile << 'EOF'
FROM qwen2.5-coder:14b
PARAMETER num_ctx 65536
EOF

# Build the extended-context model
ollama create qwen-cline -f qwen-cline-modelfile

# In Cline settings (UI)
# API Provider: Ollama
# Base URL: http://localhost:11434
# Model ID: qwen-cline
# Save

# For better performance, enable Compact Prompt
# Cline Settings → Features → Use Compact Prompt → ON
# Reduces system prompt size by ~90%

# Add to ~/.zshrc or ~/.bashrc to persist
export OLLAMA_KEEP_ALIVE=24h     # keep model loaded all day
export OLLAMA_FLASH_ATTENTION=1  # enable flash attention
export OLLAMA_KV_CACHE_TYPE=q8_0 # 8-bit KV cache reduces VRAM usage
export OLLAMA_CONTEXT_LENGTH=8192 # or higher if VRAM permits

# Restart Ollama service to apply
brew services restart ollama  # macOS
# or: systemctl --user restart ollama  # Linux systemd

# Set persistent user env vars
[Environment]::SetEnvironmentVariable("OLLAMA_KEEP_ALIVE", "24h", "User")
[Environment]::SetEnvironmentVariable("OLLAMA_FLASH_ATTENTION", "1", "User")
[Environment]::SetEnvironmentVariable("OLLAMA_KV_CACHE_TYPE", "q8_0", "User")
[Environment]::SetEnvironmentVariable("OLLAMA_CONTEXT_LENGTH", "8192", "User")

# Restart Ollama (reopen the Ollama app from system tray)
# Or kill and restart from CLI:
Get-Process ollama -ErrorAction SilentlyContinue | Stop-Process -Force
ollama serve