System Overview & Mission

GnamiSI (gnamiai.com) is a desktop workspace for Windows and macOS, with a Linux build in progress. It runs on your machine. Open a project, choose a model, and work on your own files. The four modes are Autopilot, Forge, Stream, and Agent.

Desktop Client

While the API, authentication, and PayPal billing reside on Cloudflare Workers, the coding client runs on Windows and macOS. A Linux build is in progress. The desktop app can read and write local files and run commands in your workspace.

# Download and run the native Windows installer
curl.exe -L -o GnamiAI-Setup-win-x64.exe https://gnamiai.live/downloads/releases/GnamiAI-Setup-win-x64.exe
./GnamiAI-Setup-win-x64.exe

Workspace Operational Modes

GnamiSI has four modes. Switch modes depending on what you need.

Time-Machine Git & Rewinds

Located directly above the code editor workbench, the Time-Machine Git timeline records micro-commits to an in-memory virtual branch prior to any mutating agent action. Regressions are rolled back with 1-click sub-second latency.

interface MicroCommit {
  id: string;
  hash: string;
  message: string;
  timestamp: number;
  author: 'agent' | 'user' | 'system';
  files: { [filePath: string]: string };
}

// Instant rollback handler
function rollbackToCommit(commitId: string): void {
  const target = virtualBranch.get(commitId);
  workspace.restore(target.files);
}

Human-in-the-Loop (HITL) Safety

Whenever an autonomous agent requests permission to execute terminal commands or overwrite critical schema files, an interruptive modal appears. In accordance with strict black-and-white design rules:

Supported Model Access

Local agent runs use a supported CLI on your machine. Cloud runs use the separate model list shown in the Cloud picker.

Cloudflare Edge & Login Rate Limiter

All website assets and API routes are deployed behind Cloudflare Workers. Authentication endpoints are safeguarded by a Cloudflare Edge Rate Limiter enforcing 5 requests per 60 seconds per IP, preventing brute-force attacks and credential stuffing.

HTTP/1.1 429 Too Many Requests
Retry-After: 60
X-RateLimit-Limit: 5
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 1773574060
Content-Type: application/json

{
  "error": "Too Many Requests",
  "message": "Cloudflare Login Rate Limiter: Too many login attempts. Please try again in 60 seconds.",
  "retryAfter": 60
}

PayPal Subscriptions & Billing API

PayPal is the exclusive payment processor for recurring billing of GnamiSI Pro ($29/month). A subscription provisions 12,500 cloud credits per month. Each month the balance resets to the plan allowance: 12,500 while a Pro subscription is active, including when Lifetime is also owned, and 2,500 for Lifetime once Pro has ended. Unused credits do not roll over. Runs through your installed Codex CLI or Grok CLI never consume credits. A cloud run sends workspace file contents to the Cloudflare Worker, which calls the selected Workers AI model and returns file contents for the desktop app to write back locally. A completed run bills 20 credits and is limited to 20 runs per user per UTC day with a 120-second maximum runtime. Failed runs caused by model or provider errors do not count toward the 20-run daily limit; timeouts do. Separately, you can start at most 40 cloud runs per UTC day, and at most 5 per minute. Failed runs cost zero credits. A minimum balance of 10 credits is required to start a run. Brief chit-chat bills a 2-credit reply instead of a full run.

Scheduled Cloud Runs

Accounts with a Pro or Lifetime plan can schedule a cloud run to start automatically. Open Scheduled cloud runs in the desktop sidebar or on your Account page, then choose a prompt, a cloud model and a schedule: daily at a set time, weekly on chosen days, or a custom five-field cron expression (minute, hour, day of month, month, day of week). Each schedule keeps the time zone you pick, so a run set for 09:00 stays at 09:00 local time across daylight-saving changes. If a time does not exist because clocks spring forward, the run starts at the same offset after the change (02:30 becomes 03:30); if a time happens twice when clocks fall back, it runs once, at the first occurrence.

Schedules run on the GnamiSI cloud, so the desktop app does not need to be open. Each scheduled run is a normal cloud run with the same models, 120-second limit and credit rules described in PayPal Subscriptions & Billing, and it counts toward the daily run limit and the daily and per-minute start limits. Scheduled runs do not include workspace files. Files the model proposes are kept in run history for review and are never written to your computer automatically.

If your balance is below the minimum or a daily limit has been reached, that run is skipped and the reason is recorded in history. If your credits run out, the schedule pauses and stays paused until you resume it. A schedule also pauses if your subscription ends, its model is no longer available, or you are signed out. Resuming continues from the next upcoming time; missed runs are not replayed. A run that cannot start within 30 minutes of its scheduled time is recorded as missed.

Run history shows the status, output, credits used and start and finish times of your last 200 scheduled runs. When a scheduled run finishes or fails, or a schedule pauses, GnamiSI sends a notice to your signed-in desktop app through remote control, and MCP clients can read the latest status with the scheduled_runs tool. You can keep up to 10 schedules, and runs of the same schedule must be at least 15 minutes apart.