Cursor CLI (agent) installed on ALL 3 Quik Cloud machines on March 12, 2026.
Why: Free API providers (Groq, Cerebras, SambaNova) hit daily rate limits causing farm downtime. Cursor Ultra gives consistent, high-quality models with no rate limits on auto and composer models.
How to apply:
- Cursor API key stored in SSM:
/quik-nation/shared/CURSOR_API_KEY - CLI installed at
~/.local/bin/agenton all 3 machines - EC2s: PATH and API key auto-loaded via
.bashrc - Mac (Quik Cloud 1): PATH and API key in
.zshrc, keychain must be unlocked first - Mac keychain unlock:
security unlock-keychain -p 'password' ~/Library/Keychains/login.keychain-db - Mac login password in SSM:
/quik-nation/quik-cloud/login-password - Mac does NOT have AWS CLI — API key hardcoded in
.zshrc(from SSM on our end) - ALWAYS use
--model auto— this is the unlimited model on Ultra plan - Named models (sonnet-4.6, opus-4.6, gemini, gpt-5.x) have monthly limits — use only when specifically needed
- Command:
agent -p --trust --model auto "task prompt" - 13 free API providers (Groq, Cerebras, etc.) become BACKUP ONLY
- Custom agentic-loop.js remains as fallback if Cursor CLI is ever down
- Quik Cloud 1 (Mac M4 Pro): INSTALLED and TESTED — working as of March 12, 2026
Cursor Ultra Plan Details ($200/month)
- Billing cycle: Feb 16 - Mar 16, 2026 (resets March 16)
- Auto + Composer pool: UNLIMITED (this is what
--model autodraws from) - API / Premium models: 20x usage vs Pro plan. Current cycle: 860.7M tokens (100% used on gpt-5.3-codex), Auto+Composer at 431M tokens (38.2% used)
- After March 16 reset: Full premium model access restored — can use named models for complex tasks
Available Models (All Providers)
- Anthropic: Claude 4 Sonnet, 4.5 Haiku, 4.5 Opus, 4.5 Sonnet, 4.6 Opus, 4.6 Opus (Fast), 4.6 Sonnet
- OpenAI: GPT-5, GPT-5 Fast, GPT-5 Mini, GPT-5-Codex, GPT-5.1/5.2/5.3 Codex, GPT-5.4
- Google: Gemini 2.5 Flash, 3 Flash, 3 Pro, 3.1 Pro
- Cursor: Composer 1, Composer 1.5
- Other: Grok Code (xAI), Kimi K2.5 (Moonshot)
Cloud Agents (Revenue Opportunity)
- Cloud agents run in isolated cloud environments — NO local machine needed
- Run as many agents in parallel as needed — infinite concurrency
- Clone repo from GitHub/GitLab, work on separate branch, push changes
- Support MCP servers for database/API/third-party access
- Use cases: test suites, builds, deployments, data processing — all billable per task-hour
- Access from: cursor.com/agents, Desktop, Slack, GitHub, Linear, API
- Billed at API rates for selected model
- This is a revenue stream: charge clients per task-hour for long-running cloud agent work