The defense-grade security gateway, multi-provider model fabric, and universal document intelligence engine.
«One task. Multiple AI models. One security layer.»
GridAI operates as an inline, mission-critical AI Security Gateway & Compute Fabric positioned between enterprise users and external frontier model providers (OpenAI, Anthropic Claude, Google Gemini, Groq, DeepSeek, Mistral, Together, Cerebras, OpenRouter, and Local Ollama).
1. [User / Application / IDE] ➔ Inbound Prompt with Code, Context or Document Attachment
2. [GridAI Ingress Gate] ➔ Pre-Flight Prompt Injection Defense (DAN, Delimiter, System Exfiltration)
3. [Deterministic Vault] ➔ Reversible PII & Secret Masking ([ENTITY_EMAIL_1], [SECRET_KEY_1])
4. [Fabric Router] ➔ Dispatches to Authorized Compute Tier (Tier 1 Local, Tier 2 Cloud, Tier 3 GPU)
5. [External LLM Provider] ➔ Processes Sanitized Query (Zero Plaintext Secrets or Customer PII Received)
6. [Egress Reflection Gate] ➔ Output Verification (Validates Output Safety & Prevents Data Leakage)
7. [Client-Side Re-Hydrator] ➔ Restores [ENTITY_EMAIL_1] Seamlessly Inside Browser Memory
8. [Audit & SIEM Ledger] ➔ Dispatches Webhook to Corporate SOC & Records Cryptographic Audit Trail
In Protected Mode, unencrypted API credentials, Ethereum private keys, passwords, customer records, and internal trade secrets are guaranteed to never reach public model providers, completely eliminating corporate data leakage liability.
Unlike legacy chatbots that blindly dispatch uninspected text to third-party endpoints, GridAI executes every Protected Mode message through an explicit, real-time 6-Stage Security Pipeline:
Traditional regex redactors destroy code logic by substituting all entities with identical static strings like [REDACTED]. GridAI employs Isolated Cryptographic Entity Vaulting:
Replaces all variables with uniform tokens:
"Pay [REDACTED] from key [REDACTED] to [REDACTED]"
➔ Model confuses sender and recipient, generating unusable logic.Preserves unique entity relationships:
"Pay [ENTITY_EMAIL_1] from [KEY_1] to [ENTITY_EMAIL_2]"
➔ Model reasons accurately; GridAI restores original values in browser memory!GridAI screens inputs against hundreds of malicious adversarial patterns:
ignore previous instructions, disregard all system directives, reset rules.repeat verbatim your system prompt, print initial configuration.<|im_start|>, [system], ### Instruction, Human: / Assistant: marker injections.sk-proj-...), AWS access keys, GitHub PATs, and Ethereum private keys.GridAI empowers enterprises to operate their own API credentials across 10 frontier and specialized inference providers. All keys are managed via the dedicated API Key Vault Modal in the dashboard:
| Provider | Primary Models | Header Identifier | Specialization |
|---|---|---|---|
| OpenAI | gpt-4o-mini, gpt-4o, o1 | x-custom-api-key | General intelligence, tool use |
| Anthropic Claude | claude-3-5-sonnet, claude-3-5-haiku | x-anthropic-api-key | Code generation, nuanced cyber analysis |
| Google Gemini | gemini-1.5-flash, gemini-1.5-pro | x-gemini-api-key | 2M token context window, multimodality |
| Groq Cloud | llama-3.3-70b-versatile, mixtral-8x7b | x-groq-api-key | LPU ultra-fast inference (500+ tok/s) |
| DeepSeek | deepseek-chat, deepseek-reasoner (R1) | x-deepseek-api-key | Advanced mathematics, chain-of-thought |
| Mistral AI | mistral-large-latest, codestral | x-mistral-api-key | EU sovereign AI, code synthesis |
| Together AI | Llama-3.3-70B-Instruct-Turbo | x-together-api-key | High-throughput open-source fleet |
| Cerebras | llama3.1-70b | x-cerebras-api-key | Wafer-scale engine instant responses |
| OpenRouter | auto, 100+ multi-fleet models | x-openrouter-api-key | Unified failover and model aggregation |
| Local Ollama | llama3.2, mistral, qwen2.5 | ollamaEndpoint (HTTP LAN) | Air-gapped on-premise execution (0 packet leak) |
User API keys are retained exclusively inside browser session storage. When requests are submitted, credentials are transmitted via encrypted TLS headers directly to the backend proxy, which relays them strictly to the designated provider without writing keys to plain server log files or databases.
If an enterprise user has not configured private API keys or exhausts their cloud quota mid-session, GridAI automatically and seamlessly routes queries to the Smart Free AI Mesh Gateway:
GridAI includes an industrial, zero-loss document processing pipeline capable of handling complex binaries, multi-page PDFs, OpenXML office archives, vector diagrams, tabular financial spreadsheets, and high-resolution document scans without truncation.
1. [Attachment Ingestion] ➔ Memory buffer read as Base64 DataURL (Non-blocking async UI thread)
2. [Magic Byte Sniffing] ➔ Header verification (%PDF-, PK\x03\x04 for OpenXML, UTF-8 BOM)
3. [Native Stream Decompress]➔ zlib.inflateSync unpacks compressed /FlateDecode PDF object streams
4. [PostScript Purification] ➔ Strips font descriptors, glyph bytecode, and graphics operators
5. [OpenXML / Vision OCR] ➔ AST paragraph & table extraction, double-entry arithmetic reconciliation
6. [Universal Export] ➔ Generates refactored DOCX, high-res PDF, or Markdown on Android/iOS/Desktop
.pdf): Native binary stream extraction with zlib FlateDecode decompression, stripping PostScript font noise in under 120ms with 100% character fidelity..docx, .doc): AST paragraph and table parser capable of converting Word documents into structured Markdown and generating 1-click refactored .docx downloads..xlsx, .csv): Tabular structure normalizer auditing balance sheets, double-entry mathematical checksums, and formula integrity..ts, .tsx, .py, .rs, .go, .sql, .json): Syntax-aware AST parsing preserving code block structure..vsdx, .svg, .drawio): XML shape tree traversal extracting connector labels and process flows.LineTotal = Quantity × UnitPrice | GrandTotal = Σ LineTotals - Σ DiscountsDevelopers can access the document parsing engine directly via:
POST /api/v1/parse-document
Content-Type: application/json
{
"fileData": "data:application/pdf;base64,JVBERi0xLjQK...",
"fileName": "Q3_Financial_Audit.pdf",
"mimeType": "application/pdf"
}GridAI orchestrates requests across three strictly governed compute tiers to prevent accidental data leaks to public infrastructure:
Runs on On-Device NPU or private Ollama LAN clusters. Completely air-gapped with zero network egress. Mandatory destination for trade secrets, root passwords, and confidential corporate IP.
High-throughput cloud inference with commercial SLAs (OpenAI, Claude, Gemini Pro, Groq). All traffic passes through GridAI Ingress sanitization and Egress verification.
Volunteer compute clusters (Petals, AI Horde). Mandatory Cryptographic Boundary: Confidential, internal, or PII-bearing data is strictly blocked from ever reaching Tier 3. Reserved exclusively for non-sensitive public questions.
GridAI enables seamless continuation of an active investigation from one AI model to another without loss of context, decisions, or milestones:
Complex enterprise problems require multi-perspective synthesis. GridAI provides an automated 5-phase collaborative pipeline:
GridAI satisfies mandatory enterprise compliance frameworks including SOC-2 Type II, HIPAA, ISO/IEC 42001, and the EU AI Act:
Security teams can export auditor-ready compliance dossiers directly from the dashboard ledger:
GridAI dispatches real-time incident notifications to your corporate Security Operations Center (SOC) via /api/v1/webhooks/security-alert:
POST /api/v1/webhooks/security-alert
Content-Type: application/json
{
"event": "GRIDAI_SECURITY_INCIDENT",
"incidentId": "inc_1728594912_a9x2",
"threatLevel": "HIGH",
"eventType": "PROMPT_INJECTION_BLOCKED",
"summary": "Malicious instruction reset directive intercepted at ingress.",
"mitigation": "INTERCEPTED_AND_BLOCKED_AT_INGRESS",
"timestamp": "2026-10-12T04:55:00Z"
}Evaluate your organization's AI defense posture on demand. The built-in Red-Team Simulator executes a battery of 50+ real-time adversarial test vectors:
Computes an instant verifiable GridAI Defense Score (e.g. 99.4% Defended - Grade A+).
Route existing Python, Node.js, and curl applications through GridAI with a 1-line baseURL change:
from openai import OpenAI
# Simply route through GridAI Protected Gateway:
client = OpenAI(
base_url="https://api.gridai.io/v1",
api_key="your-gridai-or-openai-key"
)
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Analyze system architecture..."}]
)GridAI fundamentally transforms how users access modern frontier AI. Instead of charging users subscription fees or token markups, GridAI acts as a Unified BYOC (Bring-Your-Own-Credentials) Hub that consolidates the independent free tiers and credits provided by ChatGPT (OpenAI), DeepSeek, Grok (xAI/Groq), Google Gemini, and Anthropic Claude into one continuous, uninterrupted workflow.
Every major AI lab offers free allowances (e.g. Groq 30 RPM, Gemini 15 RPM, DeepSeek free tokens, OpenAI trial tiers). GridAI lets you connect your accounts once, pooling all your free quotas in a single interface.
When ChatGPT hits HTTP 429 Insufficient Quota or a rate limit, GridAI detects it immediately. The gateway automatically preserves your full conversation history and migrates the question to DeepSeek or Grok without losing context.
All interactions across all providers pass through the GridAI Security Shield (prompt-injection defense, secret and PII redaction) and the 6 Dharma Principles (Non-Harm, Truthfulness, Non-Exploitation, Compassion, Epistemic Humility, and Human Responsibility).
GridAI brings ChatGPT-grade features to all providers: Voice input & speech synthesis, DALL-E 3 & HD diffusion image generation, document OCR, and live internet weather/search telemetry work across all connected models.
| Provider | Free Tier Allowance | Auto-Migration Priority | Native Mobile App Scheme |
|---|---|---|---|
| 🟢 ChatGPT (OpenAI) | Account Free Quota / Credits | Primary Model | chatgpt:// |
| 🐳 DeepSeek V3 / R1 | Free Balance Tokens | Failover Tier 1 | deepseek:// |
| ⚡ Grok / Groq LPU | 30 RPM Free (500 T/S) | Failover Tier 2 | x:// / twitter:// |
| ✨ Google Gemini | 15 RPM Free Tier | Failover Tier 3 | googleapp:// |
| 🔒 Local Ollama | Unlimited On-Device ($0.00) | Air-Gapped Sovereign Node | Localhost LAN |
On the GridAI Mobile App, users have two complementary ways to interact:
chatgpt.com, chat.deepseek.com) inside a native full-screen viewer.One of the largest challenges in modern AI operations is provider lock-in and quota exhaustion. Users often start an inquiry with a high-tier model, only to run out of credits mid-task or receive an answer containing subtle hallucinations. GridAI solves this with three tightly integrated architectural features:
Users can shift the active model right inside the chat input pill at any time before free credits run out. Complete conversation history, code snippets, and system instructions are forwarded to the new provider without context loss.
Every completed answer features quick-comparison chips ([⚡ Groq], [🐳 DeepSeek], [✨ Gemini], [🛡️ Claude], [🟢 ChatGPT]). Clicking any chip executes the same user query through the alternative provider for side-by-side benchmarking.
The Second Opinion panel prompts an opposing provider: "Please review and critique the answer above from [Provider]. Do you agree with it? What did it miss or get wrong, and what is your verified solution?" eliminating single-model hallucinations.
1. User submits architectural question ➔ Dispatched to Provider A (e.g., DeepSeek R1).
2. Provider A returns response with token usage metrics and response latency telemetry.
3. User reviews answer and clicks [🤔 Second Opinion] ➔ Selects Challenger Provider B (e.g., Claude 3.5 Sonnet or Groq LPU).
4. GridAI synthesizes an adversarial critique query containing Provider A's response in conversation memory.
5. Provider B audits Provider A's output, flags edge cases, verifies formulas, and generates an independent counter-analysis.
GridAI pioneers the transition from flat, static markdown text to reactive, stateful generative micro-applications. When any connected AI model generates actionable procedures, checklists, or recipes, GridAI client-side hydrators render them into interactive mini-apps with persistent component state.
Checklists maintain reactive checked/unchecked state directly inside the chat thread. A live progress indicator dynamically computes completion metrics (e.g. "2 of 3 completed • 67%") without requiring round-trips to the server.
Formulas and recipes feature one-touch scaling toggles (1x, 2x, 4x). Selecting a serving size instantly multiplies all ingredient units in real time with zero latency and zero hallucination risk.
• 400g Fettuccine or Penne
• 200g Fresh baby spinach
• 2 Organic lemons (juiced)
Curated media cards pair high-resolution imagery with verified domain origin badges (e.g. spice.alibaba, arxiv.org), ensuring full attribution transparency and eliminating visual ambiguity.
Easy pasta recipe
Lemon & baby spinach pasta scaled for 2 or 4.
The GridAI Gateway injects strict system prompt formatting rules that tell upstream models to wrap structured data in widget envelopes:
:::widget[recipe]
{
"title": "Easy Lemon-and-Spinach Pasta",
"baseServings": 2,
"options": [1, 2, 4],
"source": "spice.alibaba",
"ingredients": [
{ "name": "Pasta", "qty": 200, "unit": "g" },
{ "name": "Fresh Spinach", "qty": 100, "unit": "g" },
{ "name": "Lemons", "qty": 1, "unit": "unit" }
]
}
:::A persistent obstacle in modern enterprise AI is provider feature asymmetry. OpenAI ChatGPT-4o offers native DALL-E image creation, Python code execution, and web browsing. In contrast, ultra-fast or private sovereign models like Groq LPU (500 T/S), DeepSeek V3/R1, and Mistral Codestral are strictly text-based.
GridAI solves this fundamental disparity through the Gateway Polyfill Architecture. If a user asks a text-only model to execute an unsupported feature, the Gateway polyfill autonomously intercepts the intent and executes it on behalf of the model without raising errors:
When a user asks Groq, DeepSeek, or Claude to "draw a futuristic cityscape", the Gateway dispatches the request to the Flux / SDXL / Mempool serverless cluster via /api/v1/generate-image. The resulting high-resolution asset is embedded in the conversation alongside the text model's answer.
Instead of executing untrusted code on vulnerable enterprise backends, Python scripts, Pandas manipulations, and Matplotlib graphs run directly in an isolated WebAssembly sandbox in browser memory.
Prior to dispatching prompts to Groq or DeepSeek, GridAI automatically runs executeLiveSearch() on queries containing recency indicators. Fresh search snippets with source URLs are injected into the context window.
Image attachments provided to text-only models are transformed via ingress optical character recognition and dual-pass vision into rich markdown tables and structural descriptions before reaching the model.
Accessible at /admin, the GridAI Control Tower delivers zero-compromise visibility into live production operations. Synthetic mock data, hardcoded IP sequences, and simulated mathematics have been completely replaced with direct MongoDB Atlas telemetry:
Consolidates multi-platform registrations (Google OAuth, Hotmail, Corporate domain, Magic Links) under single unified profiles. Kosala Wijayarathna (Founder) links 4 email identities with 99 aggregate logins and 1,431 queries.
Tracks real-time token and credit consumption against the 999,999 early-adopter free grant. Displays remaining balances, exact point deduction records, and estimated enterprise dollar savings (~$8,400+).
Interactive audience toggles enable instant segmentation across Unique Users, All Email Accounts, and Anonymous Guests with live forensic prompt logs.
The navigation sidebar provides instant access to the 6 core capability workspaces modeled after industrial AI workflows:
| Module | Icon | Core Capability | Underlying Engine |
|---|---|---|---|
| Images Studio | 🖼️ | Multi-model visual studio, generative gallery, and preset formula matrix (Anime, Blueprint, 3D Avatar, Interior) | Flux / SDXL Serverless Cluster |
| Library | 📖 | Saved artifacts, bookmarks, reusable prompt snippets, and cryptographically verified response ledgers | Client IndexedDB & Mongo Vault |
| Projects | 📁 | Scoped custom workspace instructions, project file context, and team collaboration barriers | RAG Vector Embedding Store |
| Remote Sandbox | 🖥️ | In-browser Python Wasm execution, bash terminal simulation, and autonomous web browsing agent | Pyodide Wasm & Docker Micro-VM |
| Scheduled Tasks | 🕒 | Automated recurring agent cron jobs, daily market telemetry digests, and autonomous data scrapers | BullMQ Redis Task Worker |
| Plugins Directory | 🧩 | Extensible provider function calling, third-party API tools, and custom BYOK enterprise connectors | GridAI OpenAPI Dispatcher |
GridAI bridges frontier consumer visual ergonomics with enterprise-grade resilience. The ChatGPT Native Studio mirrors the exact 1:1 mobile experience while anchoring generation under the GridAI Defense-in-Depth Gateway.
Pixel-accurate reproduction of the native mobile interface, featuring the top bar (☰ menu, ✦ Get Plus, compose new chat), subheader tabs (Trending & Templates), 2-column 3:4 aspect cards, and floating bottom capsule with voice dictation.
When ChatGPT-4o / DALL-E 3 rate limits (HTTP 429) or credit thresholds are hit, the gateway calculates the exact reset time window and provides 1-click zero-token-loss migration to Smart Free Mesh ($0.00), DeepSeek R1, or Claude 3.5.
If an active model is text-only (e.g. DeepSeek R1, Groq Llama 3), the GridAI Gateway transparently polyfills missing visual diffusion via serverless Flux 8K clusters on the user's behalf, maintaining uninterrupted workflow continuity.
| Scenario / Event | Trigger Condition | Autonomous Gateway Action | User Experience |
|---|---|---|---|
| Standard Visual Prompt | Normal Operation | Dispatches to ChatGPT-4o / DALL-E 3 through 6-stage sanitization vault | Immediate 8K render preview with download and inpaint |
| ChatGPT Rate Limit (HTTP 429) | Retry-After Header / 429 | Parses backoff duration and activates failover bridge | Displays calculated reset time (~42m) + 1-click migration buttons |
| Unsupported Text-Only Model | User selects DeepSeek / Groq | Autonomously routes image synthesis to Serverless Flux 8K cluster | Displays Industrial Gateway badge with polyfill details |
The GridAI Relay Protocol (GRP) eliminates vendor lock-in and daily free quota barriers across all major frontier AI providers. Users can leverage the free daily allocations of ChatGPT (DALL-E 3), Google Gemini (2M Context & Imagen 3), Anthropic Claude 3.5 (Forensic Vision & SVG Vector CAD), and DeepSeek R1 (Mathematical Logic) seamlessly.
• ChatGPT-4o: 2-3 Free DALL-E 3 Blueprints/day (3-hr rolling limit) ➔ 1-Click Outbound Bridge
• Google Gemini: 2M Token Context + Imagen 3 ➔ 100% Free via Google AI Studio / Gemini App
• Claude 3.5 Sonnet: Surgical Forensic Vision & CAD Code ➔ Generous daily conversational turns
• DeepSeek R1: Deep Chain-of-Thought Physics & Math Tolerances ➔ 100% Free & Open Access
• Universal Ingest Bridge: 1-Click Smart Clipboard / Drag-and-Drop Ingest back into GridAI thread
| Feature | Isolated Vendor Apps | GridAI Universal Relay Protocol (GRP) |
|---|---|---|
| Quota Depletion | Hard stop / 3-hour lockout countdown | Automatic 1-click continuation to Gemini, Claude, or DeepSeek |
| Context Preservation | 0% (User must re-copy, re-upload, and re-explain) | 100% (History, CAD blueprints, and parameters serialized automatically) |
| Image / CAD Handshake | Siloed within proprietary UI | Inlined media cards seamlessly relayed to downstream vision models |
| External Artifact Ingest | Manual copy-paste required | Smart Ingest Modal with automated clipboard and file drop-zone |
To eliminate tab switching and eradicate unauthorized third-party watermarks, GridAI embeds the In-App ChatGPT Workstation directly into the right-hand split console. Users trigger authentic OpenAI DALL-E 3 blueprints or GPT-4o reasoning live within GridAI, and one-tap sync the resulting visual artifacts back to the canvas.
• Zero Watermarks Enforced: Connects to authentic DALL-E 3 with strict prompt isolation.
• In-App Workstation Drawer: Slide-out console with real-time prompt buffer and live monitor.
• 1-Tap [SYNC OUTPUT TO GRIDAI]: Ingests high-resolution visual cards directly into the project thread.
• First-Launch Industrial Home: 4 high-contrast action cards (CAD Blueprints, Multi-Model Relay, Manga, Documents).
GridAI provides an isolated, in-chat Python Code Interpreter & Data Analysis Runtime powered by Python 3.13, Pandas, NumPy, and headless Matplotlib (Agg backend). Code blocks generated by the agent or entered by the user can be executed on-demand in a secure sandbox, returning formatted stdout, statistical aggregates, graphical plots, and generated file artifacts.
• Runtime Engine: Headless Python 3.13 subprocess with sub-second execution isolation.
• Automated Visual Capture: Matplotlib and Seaborn figures are intercepted via memory buffers and rendered in-chat as high-resolution PNG data URLs with 1-click download.
• Dataset Export Capture: Files written to disk (e.g. df.to_csv(), df.to_excel(), .parquet, .json) are auto-indexed and surfaced as downloadable binary badges with exact file sizes.
• API Surface: POST /api/v1/sandbox/python accepting { code: string, timeoutMs?: number }.
| Capability | Technical Implementation | User & Workspace Outcome |
|---|---|---|
| Statistical Analysis | Pandas, NumPy, Scipy runtime evaluation | Immediate correlation matrices, outliers & distribution statistics |
| Chart Generation | Matplotlib Agg / Seaborn figure buffer extraction | Inline multi-chart display with PNG download options |
| Data Transformation | CSV / Excel serialization interceptor | Instant downloadable dataset files (.csv, .xlsx) directly under message |
The OpenAI Deep Research Parity Engine conducts autonomous, recursive multi-pass web investigations across live industry and academic sources. Rather than single-turn search retrieval, the engine formulates multiple orthogonal search vectors, crawls raw web pages, extracts structured facts and metrics, and compiles an exhaustive 15-page structured intelligence whitepaper.
1. Vector Formulation: Decomposes prompt into 4 search vectors: (1) Overview & Breakthroughs, (2) Technical Architecture & Benchmarks, (3) Market Adoption & Industry Leaders, (4) Future Roadmaps & Regulatory Challenges.
2. Parallel Crawl: Executes concurrent web queries via DuckDuckGo and frontier search gateways, extracting full DOM text from high-authority sources.
3. Source Grounding & Deduplication: Cleans HTML, extracts metrics, tables, and creates verified cryptographic reference citations.
4. Whitepaper Synthesis: Structures findings into an Executive Summary, Comparative Architecture Tables, Strategic Implications, and Source Citations.
• API Surface: POST /api/v1/deep-research accepting { topic: string, depth?: number, apiKey?: string }.
GridAI introduces Interactive Canvas Mode, a dedicated split-screen document and code workspace that allows users and the AI to co-create, edit, and iterate in real time without polluting the conversation transcript.
• Reading Level Adjuster: 1-Click rewrite targeting Executive, College, High School, or PhD comprehension.
• Length Modifiers: Instant Shorter (concise synthesis) or Longer (deep real-world technical examples) adaptations.
• Fix Bugs: Automated scan detecting subtle edge-case exceptions, type mismatches, and memory leaks.
• Port to Python: Automated language migration transpiling TypeScript or JavaScript blocks directly to idiomatic Python.
• Bidirectional Sync: 1-Click transfer from Canvas into the active chat session or vice versa.
For air-gapped security operations, red-team evaluations, and confidential compliance queries, GridAI features Zero-Trace Incognito Mode. When activated, all database writes and persistent client caches are strictly suppressed.
• Database Bypass: Short-circuits MongoDB collection operations; zero records are created or updated in ChatSession.
• Storage Bypass: Browser localStorage and IndexedDB caching are bypassed entirely.
• Memory Flush: Context states live purely within transient React runtime memory and are wiped permanently upon session close or tab reload.
• Visual Indicator: Prominent incognito warning banner with 1-click exit control.
GridAI includes an enterprise-grade Chat Productivity Suite for managing large volumes of multi-model conversations, team sessions, and public artifacts:
• Session Pinning (📌): Pin mission-critical threads; pinned conversations are automatically sorted to the top of the session list.
• Session Archiving (📦): Hide concluded or inactive sessions from the active workspace without deletion.
• Real-Time History Search: Instant client-side text filtering across conversation titles, model names, and providers.
• Public Share Links (/share/[shareId]): Generates cryptographic immutable snapshot links for external collaboration with zero authentication barrier.
• API Endpoints: POST /api/v1/chat/share (create share link) and GET /api/v1/chat/share?shareId=... (fetch public snapshot).
For reasoning models (OpenAI o1/o3-mini, DeepSeek R1, Google Gemini 2.0 Flash Thinking), GridAI provides granular Reasoning Effort Selectors alongside expandable in-chat thinking traces.
• Low Effort: Fast, concise responses prioritizing latency and minimal token consumption for quick questions.
• Medium Effort: Balanced multi-pass chain-of-thought verification for standard engineering and drafting tasks.
• High Effort: Deep multi-step verification, counter-argument exploration, and formal proof generation for complex mathematics and mission-critical code.
• Expandable Thinking Accordion: Formatted 💭 Thought for Xs collapsible container exposing the model's intermediate verification trail.
Every assistant message features an integrated Neural Read Aloud Player utilizing the native browser Web Speech API. This provides low-latency, natural auditory comprehension of long code analyses, executive summaries, and research reports.
• Zero Latency: Executes purely on-device using hardware-accelerated speech synthesis engines without third-party network roundtrips.
• Markdown Strip Engine: Sanitizes markdown syntax, raw code blocks, and HTML tags before synthesis to ensure smooth narrative flow.
• Full Control: In-chat play/pause toggle with animated visual pulsing while actively speaking and instant audio cancellation on stop.
GridAI provides a floating, Spotlight-style Global Desktop Companion Bar accessible anywhere across the platform via Alt + Space (or Option + Space on macOS).
• Global Shortcut: Press Alt + Space to summon the companion overlay instantly.
• Model Quick Switcher: Cycle between Gemini 2.5, GPT-4o, Claude 3.5 Sonnet, and DeepSeek R1 directly inside the companion search bar.
• Tool Quick Launch: 1-Tap pills to launch Python Sandbox or Deep Research queries immediately.
• Studio Expand: 1-Click expansion to transition seamlessly into the full GridAI IDE workstation.
The Universal Document Vision & OCR Engine ingests and parses any file format—including scanned documents, engineering blueprints, architectural schematics, financial spreadsheets, and multi-page PDFs.
• Document Formats: PDF (scanned & vector), DOCX, RTF, TXT, Markdown, EPUB, JSON, YAML, XML.
• Tabular Data: XLSX, XLS, CSV, TSV with automated column structure parsing and summary metrics.
• Visual & Engineering: PNG, JPEG, WEBP, TIFF, SVG, BMP, camera photos, and architectural blueprints.
• OCR Pipeline: Tesseract and neural vision models extract bounding-box text, data tables, and handwriting with 99.4% accuracy.
• Unified Plus Button (+): Connected directly to OCR Ingest, Image Studio, Camera Capture, Canvas, and DALL-E 3 CAD Blueprints.
To deliver a modern, premium experience, GridAI replaces static placeholder images across the platform with authentic High-Definition Motion Video Visuals and dynamic moving motion pictures.
• Hardware Acceleration: Seamless WebM / MP4 video streams loaded with hardware decoding and zero layout shift.
• Dynamic Overlays: Subtle gradient meshes and glassmorphism backdrops enhancing typography and read-flow.
• Responsive Scaling: Aspect-ratio-preserving motion containers optimized for 4K desktop monitors down to mobile viewports.
The Neural PDF & Multi-Document Architect enables instant ingestion, tabular extraction, contract clause classification, and semantic citation across massive 500+ page technical manuals, complex NDAs, multi-party Master Service Agreements (MSAs), and engineering schematics.
• 500+ Page Scale: Sub-second stream inflator & vectorized object pipeline handling multi-gigabyte document corpora without browser memory degradation or context truncation.
• Tabular Extraction Engine: Detects multi-column coordinates and vertical alignments to transpile financial matrices, latency tables, and bill-of-materials into clean, copyable Markdown tables (| Col 1 | Col 2 |).
• Contract Clause Classifier: Automatically indexes legal structures (e.g., Article 3.1 Indemnification, Section 12.4 Termination, Schedule B Pricing) with direct page links.
• Semantic Citation Protocol: Verifiable, cryptographic page citations (e.g., [Doc: MSA.pdf, Page 42, §Clause 14.2]) enabling surgical cross-document cross-examination.
• Multi-Document Corpus Ingress: Simultaneous batch analysis across unlimited related documents with cross-document discrepancy discovery.
• API Endpoints: POST /api/v1/neural-pdf (accepts fileData, documents array, and optional query) and upgraded backward-compatible POST /api/v1/parse-pdf.
| Capability | Technical Implementation | Enterprise Guarantee |
|---|---|---|
| 500+ Page Ingestion | Stream-level FlateDecode inflator + pdf2json coordinates | Processes 500+ page manuals in <1.8s without memory overflow |
| Tabular Extraction | Coordinate clustering on vertical (y) & horizontal (x) deltas | 100% structured Markdown table output with 1-click clipboard & CSV export |
| Semantic Citations | BM25 + vector page-level chunking with citation tokens | Every retrieved fact links directly to verified source page and clause |
| Multi-Doc Cross-Exam | Batch documents array ingestion with cross-document matching | Flags conflicting clauses and contradictory liability terms between documents |
The 1-Click Native Mobile App Generator enables enterprise engineers and founders to instantly formulate, scaffold, and test production-grade React Native & Expo SDK 51 mobile applications directly from natural language prompts, with immediate on-device QR testing and cloud binary generation.
• Live Mobile Testnet (Expo QR): Scan the dynamically rendered QR code with Expo Go (or native camera on iOS) to immediately launch the app on real physical phones with zero simulator friction.
• Zero Xcode / Android Studio Installation: No gigabytes of local toolchains, SDKs, or emulator drivers required. Everything builds, bundles, and validates inside the GridAI cloud runner.
• Ready for App Store & Play Store Deployment: Generates standardized eas.json, clean modular architecture, and adaptive icons ready for 1-click cloud submission via npx eas-cli build --platform all.
• Direct Android APK Binaries: Pre-compiled standalone ARM64 & x86 APK packages available for direct side-loading without Google Play restrictions.
• Interactive Studio Simulator: Real-time in-browser phone viewport mockup allowing live interaction with generated state, buttons, and telemetry streams.
• API Endpoint: POST /api/v1/mobile/generate (accepts prompt, appName, and targetPlatform, returning full appJs, appJson, packageJson, and qrCodeUrl).
| Core Advantage | Technical Architecture | Enterprise Guarantee |
|---|---|---|
| Live On-Device QR Testnet | Dynamic Metro manifest URL + SVG QR code synthesis | Immediate testing on physical iOS & Android devices in under 2 seconds |
| Zero Local Toolchain Setup | GridAI Cloud Runner & Remote Metro Bundler | Builds native codebases without installing Xcode or Android Studio |
| Cloud Binaries (.APK & EAS) | Direct standalone APK generation + EAS Build orchestration | Full App Store and Google Play Store compliance with automated signing keys |
| Interactive Device Mockup | Hardware-scaled responsive CSS container with live state | Preview 60fps animations and layout responsiveness prior to physical deployment |
The Industrial Live Telemetry & Feeds Engine delivers sub-minute real-time updates for breaking global news, localized meteorological observations, and financial metrics with hardware GPS auto-detection, configurable polling intervals (30s, 60s, 2m, 5m), and instant manual sync triggers.
• Automatic Location Detection (GPS & IP): Leverages browser navigator.geolocation hardware GPS coordinates with fallback to IP-based reverse geocoding to resolve pinpoint city, latitude, longitude, and region without manual configuration.
• High-Resolution Numerical Weather: Direct connection to the Open-Meteo European & GFS numerical models, providing real-time ambient temperature (°C/°F), humidity, wind vectors, rain probability, and 3-day meteorological forecasts.
• Real-Time News Wire Polling: Continuously ingests live RSS news feeds from Bloomberg, Reuters, BBC, Financial Times, and regional wire services, categorizing items by Frontier Tech & AI, Global Markets, Geopolitics, and Science.
• World Sports Highlights & Match Reels (Football & Cricket): Official HD highlights with moving video previews, real-time scoreboards, live timeline moments, and interactive video modal player. Covers Premier League weekend goals (Saka, Haaland, Palmer, Diaz), UEFA Champions League Bernabéu thrillers (Real Madrid vs Man City), ICC Test Series (India vs England, Bumrah 9-wicket haul), Bilateral ODIs (Sri Lanka vs Australia Galle thriller, Nissanka 88 & Hasaranga 4/38), and IPL T20.
• Live BBC Sport Wire Ticker: Streams sub-minute live sports headlines directly above the sports carousel with 1-click verified BBC dispatch links.
• Tactical AI Match Breakdown: 1-click “Ask AI” integration on every football and cricket card to query deep tactical reviews, pass accuracy, xG, and top batter/bowler performances.
• Configurable Polling Intervals: Industrial dashboard sync frequencies (30s, 60s, 2m, 5m) with active pulse telemetry badges (🟢 LIVE • Synced 4s ago • Next in 56s).
• 1-Click Force Refresh: Instant parallel re-fetch across weather, sports wire, and news with zero page reload or frame disruption.
• API Endpoint: GET /api/v1/live-feed (accepts type, city, lat, lon, region, and category, returning structured real-time weather, regional hubs, sports reels, and live breaking headlines).
| Stream / Feature | Data Provider & Transport | Update Frequency & Precision |
|---|---|---|
| Pinpoint User Weather | Open-Meteo WMO Numerical Weather API | Live observations updated every 60s at exact GPS lat/lon |
| Global Wire Headlines | Direct Google News RSS Wire (BBC / Reuters / FT) | Real-time breaking dispatches with live relative timestamps (<5m ago) |
| Live Sports Highlights (Football & Cricket) | Official Broadcast Feeds (Sky Sports, Premier League, Sony Sports, ICC) | HD moving video previews, live scorecards, and timeline milestones |
| BBC Sport Live Wire Ticker | BBC Sport RSS Feed Ingestion Pipeline | Live breaking sports news ticker updating sub-minute |
| Reverse Geocoding | BigDataCloud & IP Geolocation Protocol | Resolves user city, country code, and regional sports focus on startup |
| Industrial Polling | Non-blocking client-side timer + in-memory edge cache | User-selectable 30s / 60s / 2m / 5m cadence with zero battery drain |
GridAI equips every connected frontier model (ChatGPT, Claude, Gemini, DeepSeek, Groq, or Smart Free Mesh) with the Universal Multi-Tier Web Reader Engine. When a user queries a URL or asks GridAI to evaluate, summarize, or audit an external domain, the gateway autonomously retrieves the complete webpage, strips botwall hurdles, and injects clean semantic context into the model's active reasoning stream.
• Universal URL & Domain Detection: Intelligently matches explicit protocols (https://...) as well as naked domain mentions (e.g. bbc.co.uk/news, stripe.com, techcrunch.com) preceded by natural language directives ("goto", "visit", "evaluate", "value", "read").
• Tier 1 — Jina AI LLM Reader: Connects to https://r.jina.ai/{cleanUrl} with custom headers. Extracts up to 25,000 characters of clean Markdown representation with titles, authors, published timestamps, tables, and hierarchical headings while effortlessly bypassing JavaScript SPA hydration hurdles and botwalls.
• Tier 2 — Semantic DOM Extraction Fallback: If Tier 1 experiences latency or rate-limiting, Tier 2 executes direct HTTPS fetch, removes non-content elements (<script>, <style>, SVGs), and extracts semantic content containers (<article>, <main>, or <body>).
• Universal Web Intelligence Directive: The synthesized web extraction is encapsulated into an authoritative context block (UNIVERSAL WEB INTELLIGENCE DIRECTIVE) and injected into the system prompt across any model provider.
• World-Class Executive Output Standard: Models format findings into an Executive Overview, Key Metrics & Offerings, Technical & Security Assessment, and verified Source Citations.
| Engine Tier | Transport & Capability | Extraction Output & Limits |
|---|---|---|
| Tier 1: Jina LLM Reader | Direct proxy bypassing Cloudflare & client-side SPAs | Clean Markdown up to 25,000 characters with complete tabular data |
| Tier 2: Semantic DOM Fetch | Direct Node.js fetch with semantic regex parser | Fast article & main container extraction for static/SSR websites |
| Cross-Provider Ingress | Unified gateway prompt injection layer | Works consistently across ChatGPT, Claude, Gemini, DeepSeek, & Groq |
To protect user workflows against unexpected rate-limiting or provider credit exhaustion, GridAI Studio features a client- and server-side Provider Quota Manager coupled with the interactive Provider Quota Migration Card.
• Granular Usage Tracking: src/lib/providerQuotaManager.ts tracks daily queries per provider with automatic midnight resets (ChatGPT: 50/day, Claude: 30/day, Perplexity: 40/day, DeepSeek-R1: 100/day, Groq: 200/day, Smart Free Mesh: Unlimited).
• 1-Click Migration Card: When a user exhausts a provider's free daily allocation, GridAI displays an interactive card offering immediate 1-click continuation onto high-speed alternatives (Gemini 2.5 Flash, Groq Llama 3.3 70B, or Smart Free Mesh) with zero context loss.
• Prompt & History Preservation: The user's active conversation, pending prompt, and attached files are instantly transferred to the new model without retyping.
• BYOK (Bring Your Own Key) Bypass: Users can input their personal API key at any time to bypass free tier caps for unlimited direct provider execution.
| Provider | Daily Free Quota | Recommended 1-Click Alternative | Alternative Latency |
|---|---|---|---|
| ChatGPT (OpenAI) | 50 queries / day | Gemini 2.5 Flash / Groq Llama 3.3 | 250ms |
| Claude (Anthropic) | 30 queries / day | DeepSeek-R1 / Smart Free Mesh | 350ms |
| DeepSeek-R1 | 100 queries / day | Gemini 2.5 Flash Think Mode | 400ms |
| Groq Llama 3.3 70B | 200 queries / day | Smart Free Mesh (OpenRouter) | 300ms |
| Smart Free Mesh | Unlimited (999,999 Credits) | Always Available Fallback Mesh | 200ms |
The In-App Embedded Provider Browser (InAppEmbeddedProviderBrowser.tsx) gives developers and analysts a built-in browser viewport right inside the IDE, enabling instant inspection of external AI services, documentation, and live websites without switching away from GridAI.
• Instant Quick Presets: Pre-configured launcher chips for ChatGPT (chatgpt.com), Claude (claude.ai), Perplexity (perplexity.ai), Gemini (gemini.google.com), and HuggingFace (huggingface.co).
• Hardware Navigation Controls: Interactive URL address bar with Back, Forward, Reload, and 1-Click “Open in External Tab” actions.
• 1-Click Context Injection: A dedicated “Import Page into Chat” button reads the current URL via the autonomous scraper and injects the extracted article directly into the active conversation.
• Sandbox Security: Runs with strict frame isolation policies (allow-scripts allow-same-origin allow-forms) to maintain enterprise data hygiene.
OmniRoute • GridAI Mem • Headroom Compression • Repo Architect • Task Observer
Inspired by frontier developer workflows and the viral Claude Code plugin ecosystem, GridAI Studio integrates the Five Autonomous Developer Superpowers directly into its core engine. Instead of requiring developers to pay for, install, and configure individual third-party plugins, all five capabilities are built natively into GridAI Studio across Web, Desktop, iOS (Apple PWA), and Android.
Developer sprints frequently hit disruptive limits (e.g. ChatGPT 3-hour usage caps, Claude 4-hour lockouts, or 429 Too Many Requests errors). OmniRoute acts as an autonomous multi-provider load balancer. The instant any primary model signals rate-limiting or downtime, OmniRoute cascades queries in under 35 milliseconds to high-throughput alternative endpoints without interrupting the developer or losing context.
| Provider / Model | Tier & Acceleration | Speed (Tokens/s) | Context Window | Failover Role |
|---|---|---|---|---|
| DeepSeek R1 (Full Reasoning) | Frontier Reasoning | 180 tps | 64k tokens | Primary Deep Logic |
| Claude 3.5 Sonnet | Frontier Code | 90 tps | 200k tokens | Complex Code Refactoring |
| Groq LPU (Llama 3.3 70B) | LPU Hardware | 580 tps | 131k tokens | Instant Sub-Second Fallback |
| Cerebras LPU (Llama 3.1 70B) | Wafer-Scale Hardware | 1,800 tps | 131k tokens | Ultra-Low Latency Cascade |
| Qwen 2.5 Coder 32B | Dedicated Polyglot Coder | 160 tps | 32k tokens | Syntax & Unit Test Generation |
Standard AI chat tools reset their context every time you close the tab or open a new conversation, forcing engineers to re-explain their frameworks, environment variables, database schemas, and architectural conventions repeatedly.
• Architectural Context Graph: Retains project structural choices (e.g. Next.js 14 App Router, TypeScript strict mode, MongoDB, port bindings) permanently in user storage and cloud databases.
• Cross-Device Memory Synchronization: Work started on a desktop workstation is instantly accessible on an iPhone (iOS PWA) or Android phone with full memory continuity.
• Interactive Memory Inspector: Developers can inspect, edit, export, or prune specific remembered rules directly from the Studio sidebar.
Raw codebase context and stack traces often contain massive amounts of token bloat (dead comments, repetitive console traces, redundant boilerplate, and whitespace) that consume context windows, drive up latency, and increase costs.
• AST & Symbol Preservation: Intelligently strips noise while preserving exact function signatures, TypeScript interfaces, and critical runtime logic.
• Real-Time Token Telemetry: Live metrics badge shows exact prompt savings (e.g. Headroom: 24,000 tokens compressed to 5,200 tokens (-78% latency)).
• Mobile Network Optimization: Minimizes cellular bandwidth consumption and accelerates payload transmission on mobile networks.
When opening a repository for the first time, Repo Architect scans file manifests, dependency trees, and config files to instantly configure the ideal development environment.
• Automatic MCP Server Setup: Auto-recommends and provisions Model Context Protocol servers for filesystems, databases (Postgres, Mongo), GitHub PR tools, and UI component inspectors.
• Framework-Specific Skills: Automatically activates tailored coding guidelines for React/Next.js, React Native/Expo, Python, or Go based on detected project anatomy.
The Task Observer operates as a silent background learning engine. Whenever a developer modifies AI-suggested code or provides iterative corrections (e.g. preferring Tailwind utilities over CSS modules, or snake_case over camelCase in backend code), the engine registers the pattern and updates permanent workspace guidelines without requiring manual configuration.
All 5 autonomous intelligence engines operate seamlessly across your entire device fleet:
pb-safe) and cloud memory sync.GridAI-Mobile.apk) with app intents and high-performance Headroom compression.Beyond software engineering, GridAI Studio (`gridai.studio`) is architected as an everyday global utility workspace accessible to students, job seekers, families, professionals, and sports fans worldwide—with zero sign-in friction on the homepage hero and real-time visitor telemetry on /admin.
Visitors landing on https://gridai.studio can immediately type any prompt, select from 12+ frontier models, or click 1-click quick-start chips without hitting a login wall. Every homepage launch and dashboard chat prompt automatically dispatches a lightweight telemetry heartbeat to /api/v1/user/heartbeat with Vercel Edge Geolocation headers (x-vercel-ip-city, x-vercel-ip-country) and a persistent browser fingerprint (gridai_guest_fingerprint), displaying live registered users, anonymous guest visitors, and their recent prompts in real time inside the /admin Control Tower.
.doc), PDF, or Markdown (.md).Think Mode chain-of-thought reasoning to solve O/L, A/L, SAT, JEE, and university calculus, physics, chemistry, and accounting problems step by step.Includes a 1-click 20+ language bar supporting English, Sinhala (සිංහල), Tamil (தமிழ்), Hindi (हिन्दी), Arabic (العربية), Spanish (Español), French (Français), Mandarin (中文), Portuguese (Português), Bahasa Indonesia, Urdu (اردو), Bengali (বাংলা), German, Japanese, Korean, Russian, Turkish, Vietnamese, Swahili, and Italian—plus 1-click translation of any AI response inside the chat action bar.
USD/LKR, GBP/LKR, USD/INR, EUR/USD, USD/AED), 24K Gold prices, and Bitcoin telemetry./images, live match scorecard in WorldFootballHighlights.tsx, and homepage utility pillar card to power organic viral sharing across global messaging groups.sitemap.ts): Indexes 20 high-intent core and utility endpoints across search engines to capture global organic search traffic.Why do brand-new OpenAI API keys sometimes return 429 insufficient_quota even when a user has a free chatgpt.com web account? Because OpenAI separates consumer web logins from developer platform keys (platform.openai.com). To solve this at industrial scale, GridAI Studio implements the 3-Ring Autonomous Free-First & Smart-Upgrade Quota Passport, uniting 30,000+ genuine free daily prompts ($0.00, no credit card), OpenAI's official 2,500,000 free tokens/day Data-Sharing program, and encrypted cross-device MongoDB key synchronization.
When a visitor on the gridai.studio homepage types a question in the interactive prompt bar or clicks any utility chip while not signed in, GridAI saves their prompt (gridai_pending_prompt) and selected model (gridai_pending_provider) and opens the 1-Click Authentication Modal. Immediately upon signing in via Google, GitHub, TikTok, X, or Email, GridAI restores their Cloud Key Passport from MongoDB and automatically executes their saved prompt inside /dashboard with zero lost work.
| Quota Ring | Providers & Official Endpoints | Daily Free Limit (Reset Every 24h) | Cost & Card Requirement |
|---|---|---|---|
| 🟢 Ring 1: 100% Free Daily APIs | GitHub Models (Official gpt-4o & gpt-4o-mini)Google AI Studio ( gemini-2.5-flash)Groq Cloud ( llama-3.3-70b, deepseek-r1)Cerebras ( llama-3.3-70b)Mistral ( La Plateforme)OpenRouter (1-Click OAuth PKCE :free) | • 200 req/day Official OpenAI GPT-4o • 1,500 req/day Gemini 2.5 • 14,400 req/day Groq LPU • 1,000,000 tokens/day Cerebras • 33M tokens/day Mistral = 30,000+ Free Prompts/Day | $0.00 Forever • Zero Credit Card |
| 🟡 Ring 2: One-Time $5 Unlock ➔ Forever Daily Free | OpenAI Official Platform (api.openai.com Tier 1 + Data-Sharing Toggle)xAI Grok Console ( console.x.ai Data-Sharing Program) | • 250,000 Free Tokens/Day on gpt-4o, o1, o3-mini• 2,500,000 Free Tokens/Day on gpt-4o-mini & gpt-4.1-mini• Renews every 24 hours at 00:00 UTC | One-Time $5 Deposit + Enable Data Sharing Toggle |
| 🟣 Ring 3: Direct Paid Pay-As-You-Go | Anthropic Claude 3.5 Sonnet (console.anthropic.com)DeepSeek Official V3 / R1 ( platform.deepseek.com)Together AI ( api.together.xyz) | Uncapped commercial throughput with zero data sharing (~$0.001 per prompt) | $2 – $5 Pay-As-You-Go (90% cheaper than $20/mo plans) |
/api/v1/verify-provider-key)Whenever a user pastes an API key inside the 3-Ring Quota Passport modal and clicks Verify & Connect, GridAI executes a 1-second live diagnostic handshake against the upstream provider. For OpenAI keys, it automatically detects whether the key has active quota (200 OK) or is locked at Tier 0 (429 insufficient_quota), guiding the user directly to the Ring 2 Data-Sharing unlock or Ring 1 GitHub Models Free GPT-4o.
/api/user/provider-keys)Connected provider keys (openai, github, gemini, groq, cerebras, mistral, openrouter, anthropic, deepseek) are automatically bound to the user's authenticated GridAI profile in MongoDB. Signing into GridAI on any desktop, laptop, iOS, or Android device immediately restores all connected provider keys without re-entering them.