GridAI Pro Code Studio Documentation

Queries real-time model registries and catalogs to dynamically detect newly published high-performance open-weight models (Gemma, Llama, Nemotron, DeepSeek).

Automatically parses deprecation headers, decommission notices, and degradation rates to prune sunset models from the routing pool before errors can reach users.

Isolates verified $0.00 pricing tiers and ranks them by maximum context window (1M > 128k > 32k tokens) and multimodal vision capabilities with an in-memory SWR cache.

Output chunk sizes dynamically adapt from 8 chars (short answers) to 120+ chars per tick (massive multi-file codebases), ensuring even 10,000-character outputs render in <1.2 seconds.

Sound synthesis is completely decoupled from rendering. The 130ms acoustic throttle guard caps audio ticks at a pleasant cadence (~7 ticks/sec) without buffer saturation or stream stalls.

Upon generation completion, playStreamCompletionChime() synthesizes a gentle C5-E5-G5 harmonic sine wave cue confirming your AI response is ready.

Powered by mammoth. Extracts clean plain text, headings, list structures, policy provisions, and table contents from Word documents without XML noise.

Powered by pdf2json and zlib stream inflation. Decodes custom CID /ToUnicode font character maps, strips TrueType font binaries (OS/2g, glyf) and ICC color profiles (HLino mntrRGB), producing 100% clean English text.

Parses Visio drawings (.vsdx, .vsd), SVG vector diagrams, and Draw.io charts. Extracts shape labels, text nodes (<Text>, <Shape>), process flow steps, and connector labels.

Full support for plain text (.txt), Markdown (.md), JSON (.json), CSV spreadsheets (.csv), log files (.log), XML, HTML, and YAML configuration files with control character sanitization.

Complete coverage from CAD Blueprints, Neon Lights, and 3D Avatars to Classical Oil, Watercolor, Cutaway Schematics, Paper Origami, Claymation, and Pop Art Stickers.

Powered by google/gemini-2.5-flash. Inspects arbitrary hand-drawn sketches (SVG vector lines), organic curves, spirals, characters, and photos to synthesize master prompts with zero hardcoded strings.

Multi-tier failover routing across Flux Pro, SDXL Turbo, Flux Realism, and Master Diffusion with strict binary image header validation (0xFF 0xD8 JPEG / 0x89 0x50 PNG).

Connect any physical folder (e.g. Virtual Machines or desktop directories) via the File System Access API or Desktop App. Code generated by GridAI writes directly to your local computer disk with zero confirmation delays.

Recursively indexes all subdirectories (e.g. Switch/, test/, Windows 11 x64/). Injects the full workspace relative path manifest and file contents into system context for comprehensive multi-folder audits.

Full native support for .txt, .py, .c, .cpp, .java, .go, .rs, .sql, .yaml, .xml, .sh, .env, .svg, and empty folder initializations.

Converts screenshots, UI wireframes, camera photos, and diagrams to Base64 Data URLs with up to 15 MB payload capacity for instant visual analysis by Gemini 2.0 Flash and GPT-4o Vision.

Uses <list_dir>, <view_file>, and <search_code> to inspect active files and AST symbols. Scoped queries (e.g. "only analyze gridai-studio") isolate targeted directories without unrelated file dumps.

Output format: GitHub Flavored Markdown (GFM) comparison matrices, status badges, and technical blueprints.

Applies surgical edits with <multi_replace_file_content> targeting non-contiguous code chunks in a single turn. Executes <verify_code> compiler diagnostics; upon error, the agent self-heals automatically.

<multi_replace_file_content path="src/app/page.tsx">
  <chunk start_line="1" end_line="10">
    <target>import OldHeader from './Old';</target>
    <replacement>import NewHeader from './New';</replacement>
  </chunk>
  <chunk start_line="150" end_line="160">
    <target>renderOldLayout()</target>
    <replacement>renderUpgradedLayout()</replacement>
  </chunk>
</multi_replace_file_content>

When requirements are underspecified, the agent pauses and renders an interactive multi-choice popup dialog with radio/checkbox cards, Skip, and Submit ↵ buttons.

• Supports single-select radio and multi-select checkbox modes.
• Blocks execution until the user selects and submits their response.
• Injects <tool_result tool="ask_question"> back into the agent loop.

Complex tasks enter Planning Mode generating <create_plan>. Opens a dedicated plan viewer with a prominent Proceed to Execution button that unlocks file modifications only upon user approval.

Runs long processes in the background with task IDs (task-xxx). Features live log streaming, reactive status tracking, and a top header ⚡ Tasks management HUD.

Full JSON-RPC 2.0 tool server implementation connecting external enterprise tool buses, GitHub bots, and IDE plugins directly into the GridAI agent execution pipeline.

POST /api/v1/mcp
{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "multi_replace_file_content",
    "arguments": { "path": "src/app/page.tsx", "chunks": [...] }
  },
  "id": 1
}

Races active frontier reasoning models in parallel to deliver sub-second response times without static template fallbacks:

• DeepSeek-Chat V3: High-speed general coding & AST analysis.
• DeepSeek-R1: Complex multi-step reasoning & architecture design.
• GPT-4o-mini: Rapid vision OCR & multimodal document processing.
• Qwen-2.5-Coder-32B: Surgical refactoring & syntax verification.

GridAI Studio incorporates the full autonomous engineering protocols of Google Antigravity, replacing passive chatbot advice with direct, end-to-end task execution:

When GridAI modifies or generates code, it seamlessly synchronizes the entire multi-cloud deployment pipeline under the hood:

• Atomic Git Database Commits: Pushes tree updates directly to kosalabtw/gridai-professional-dex:main via /api/v1/git.
• Automated Vercel Deployment: GitHub webhooks instantly trigger production builds, deploying updates live to gridai.studio/dashboard ↗.
• Automated Render Sync: Pings Render deployment hooks to hot-reload backend microservices and neural engine workers simultaneously.
• Zero Manual Terminal Burden: No manual git push or credential setup required from the user.

Every engineering solution, bug diagnosis, and refactor is deterministically structured according to institutional standards:

GridAI implements multi-tiered security barriers ensuring strict owner isolation across public web, desktop, and mobile environments:

• Owner-Guarded Admin Infrastructure: All administrative endpoints (/api/admin/*) enforce strict owner email verification (kosalabt@hotmail.com / kosala@gridai.studio) and master key validation, rejecting unauthorized visitors with 403 Forbidden.
• Mandatory Token Verification for Git Writes: Remote commit and pull request operations (/api/v1/git) strictly require user-supplied Personal Access Tokens, preventing public visitors from writing to the private repository.
• Client-Side Sandbox Isolation: Personal Access Tokens and session state reside strictly in your local device's isolated localStorage, with 0 exposure to external visitors.
• Hardware SSH Isolation: Local desktop operations execute directly through your physical machine's SSH key (~/.ssh/id_rsa), ensuring zero external network exposure.

Execute complex engineering operations with 1-click floating action chips and universal keyboard shortcuts:

💬 Interactive Inquiries Modal (ask_question)

When the agent requires clarification or architectural design approval, execution pauses and renders an interactive choice modal with (Recommended) cards, multi-select checkboxes, and custom write-in inputs.

📋 Implementation Plan & Proceed Gate (create_plan)

Architectural plans are rendered in a structured blueprint drawer with User Review Required, Proposed Changes, and Verification Plan. Features an executable "Proceed" gate before modifying code.

⚡ In-Editor Side-by-Side Monaco Diff Mode

Toggles seamlessly between single-file code editing and native side-by-side / inline Diff mode. Highlights red/green line deletions and additions with line-level change counters.

⚙️ Background Task & Daemon Manager (manage_task)

Spawns and monitors long-running support daemons, background test runners, and recurring cron schedules (schedule) with live log streaming, CPU/uptime telemetry, and kill controls.

🌐 Live Web Endpoint & OWASP Perimeter Scanner

Audits live domains and online APIs for Content-Security-Policy (CSP), Strict-Transport-Security (HSTS), X-Frame-Options clickjacking protection, wildcard CORS credential leaks, and cookie security flags (HttpOnly; Secure; SameSite).

⚡ 1-Click Autonomous Patch & GitHub Commit

For every discovered vulnerability, GridAI generates exact, production-ready code patches (e.g. Next.js security middleware, sanitization wrappers, parameterized queries). Clicking "Auto-Fix & Commit" immediately applies the patch and commits it to your GitHub repository.

Executes native host commands via PowerShell / Bash with a 15MB buffer and live streaming logs directly inside the Monaco terminal.

Open and bind any local folder (C:\Users\...) via native OS folder picker with recursive file tree hydration on launch.

Press Ctrl+I anywhere in the editor to open a floating prompt HUD for surgical in-editor code refactoring.

Features specialized aesthetic formulas including Technical Blueprint CAD, Studio Ghibli/Shinkai Anime, 3D Pixar Characters, Die-Cut Vinyl Stickers, Sepia Journal Field Notes, Lapel Badges, Retro 80s Neon Synthwave, and Figma Mobile UI Mockups.

Paint targeted mask overlays directly on generated artwork or reference sketches to add, replace, or refine specific objects without regenerating the entire composition.

Render two distinct artistic styles in parallel (e.g. Anime vs. Cyberpunk or Blueprint vs. 3D Avatar) to compare lighting, contours, and composition side-by-side.

Interactive perspective tracking reacts to mouse and touch movements, producing realistic 3D depth parallax, surface gloss, and cinematic presentation.

Configured with user-scalable=yes, maximum-scale=5.0, and CSS touch-action: pan-x pan-y pinch-zoom for fluid two-finger zooming and panning anywhere on mobile.

1-click fullscreen darkroom lightbox with dedicated ＋ / － controls, 100% fit-to-screen toggle, and up to 500% high-resolution inspection.

Quick double-tap gesture on any generated image card instantly switches between 100% overview and 200%–250% detail magnification with zero latency.

Execute Node.js, React, and Next.js applications directly in the browser via WebAssembly. Features Hot Module Replacement (HMR), device viewport emulation (Desktop, Tablet, Mobile), and 1-click shareable preview URLs.

Collaborate with teammates in real time using CRDT-synchronized state (Yjs). Includes multi-cursor tracking, shared terminal outputs, and synchronized AI code synthesis across Web and Desktop instances.