NeuriX Multi-ModelThree independent browser-local model lanes
Cortix AI
CortiX AI WEB Inference Engine
Ask about CortiX products, technology, patents, deployments, security posture, use cases and general AI topics. This conversation workspace opens directly from the homepage and uses local browser inference when available.
imageVision input · local browser inference
Planning agent run…
Process<Parsing><think><Reasoning><Agent>0 steps⌄
Execution timeline0.0s
Structured execution and reasoning summaries only. Hidden chain-of-thought is not displayed.
First load can be large. Browser cache is reused on later sessions.Vision is marked experimental because browser/VLM compatibility can vary.
CORTIX AI STUDIOBrowser-local intelligence workspace
STUDIO CONTROL CENTEROne surface for workspace, intelligence, tools, runtime and governance
v112.8.2
Loading active project and runtime state…
UI COVERAGE AUDITDetects upgraded capabilities that are missing, hidden or not discoverable
UI audit has not run yet.
CSV and Excel data are parsed in the browser. Dataset rows, statistics and analysis state remain in local IndexedDB.
No dataset selected.
0Rows
0Columns
0Numeric
0Missing
JavaScript and HTML run in restricted browser sandboxes. Python loads Pyodide on demand. TypeScript, JSON and CSS are analysis-only.
Output appears here.
Generated files are verified from real Blob bytes before CortiX marks them downloadable.
No generated files yet.
Native browser outputs: text, JSON, CSV, HTML and SVG. XLSX, PDF, DOCX, PPTX and ZIP load their real file libraries on demand. A failed library load is reported as unavailable rather than simulated.
Adaptive live web is shared by CortiX, NeuriX and MatriX. Current/online/URL requests use real retrieved sources or fail closed.
Browser code cannot literally read every arbitrary origin directly because cross-origin responses are controlled by the destination server. CortiX therefore uses direct CORS where available, Jina Reader/Search, optional SearXNG and the existing public specialist federation. It never substitutes model memory for a failed live-web request.
ToolRegistry and ConnectorBroker are local runtime surfaces. Remote success is shown only after an actual server response.
Connect MCP server
OAuth 2.1 authorization
OAuth secrets and PKCE verifier are kept in sessionStorage. CortiX does not impersonate an authorization server and does not fabricate a successful authorization.
Computer Use is restricted to registered real adapters. External browser/desktop control is unavailable until a real adapter is connected.
Durable Task 2.0 persists definitions, runs, retry state and approvals. Automatic standalone execution still requires an open page unless a real external executor is registered.
Create durable task
External execution boundary
The standalone file does not include a hidden worker or fake backend. This control registers only a real same-origin HTTPS service-worker script that implements the CortiX task message protocol.
Select a task to inspect its durable definition and state.
Create or import a project. Files stay in browser IndexedDB.
No file selected
Proposed change
Local mode uses the loaded WebLLM to generate SVG artwork entirely in the browser. Raster generation is optional and uses an explicitly configured image endpoint in consumer mode. CortiX AI WEB permits configured external endpoints. Zero-egress restrictions apply only to the separate CortiX AI ZEG enterprise product.
Generate
Raster generation and editing use only explicitly configured authorized endpoints. Tokens remain in sessionStorage. CortiX AI WEB permits configured external image endpoints; ZEG restrictions are not part of this Web runtime.
Auto orchestration activates for complex analysis, comparison, planning, architecture, research and review requests. All Stage 3 tools are read-only browser-local retrieval tools.
Agent roles6
Controlled tool registry6
Latest runIdle
No agent run yet.
LOCAL MODEL BENCHMARKMeasured on this browser and active WebLLM profile
—Score
—Avg latency
—Profile
Run the benchmark to measure reasoning, logic, coding and factual response behavior on the currently selected local model.
Settings apply to the next local WebLLM generation. Context length remains constrained by the selected browser model.
GENERATED FILES
Generated HTML, SVG, JSON, Markdown and code artifacts will appear here.
IndexedDBRAGMemoryAgentsArtifacts
CORTIX AI
01
What Cortix is.
One platform, defined from every angle that matters: for the buyer, the builder and the board.
The platform, defined
Cortix AI Studio is the AI workstation: a complete platform with chat, agents, memory, retrieval, live web access, tools and connected services. CortiX AI WEB is internet-enabled by design. CortiX AI ZEG is the separate enterprise server product for organizations that require zero-egress operation.
A new category
Cortix AI Studio is a full AI workstation rather than a thin chatbot shell. CortiX AI WEB combines browser-local inference with connected web research, external model and service providers, MCP, plugins, tools and cloud resources. CortiX AI ZEG uses the enterprise server architecture when isolation is required.
Risk and compliance
CortiX supports two distinct security boundaries. CortiX AI WEB is a connected product comparable in operating model to leading web assistants, while CortiX AI ZEG is the dedicated enterprise deployment for legal, healthcare, financial and other workloads that require zero-egress infrastructure.
Ownership
Cortix AI Studio keeps operator control over models, memory, tools, projects and system rules while allowing CortiX AI WEB to use the public internet and authorized external providers whenever those capabilities improve the result.
Parity
Cortix maintains two product boundaries from the same platform architecture. CortiX AI WEB is connected-first and may use live internet, remote APIs, external models and connected tools. CortiX AI ZEG is the separate enterprise server product with the zero-egress boundary.
Workstation heritage
Cortix AI Studio is to AI what the professional workstation was to computing: serious equipment for serious work. One cumulative platform carrying chat, multi-agent orchestration, cross-session memory, verified retrieval, voice and rendering, engineered for professionals whose output has consequences.
Said plainly
Cortix AI Studio puts a complete AI department behind one interface. CortiX AI WEB reasons, remembers, researches and builds with live connected resources; CortiX AI ZEG provides the separate isolated enterprise deployment.
In one line
Connected intelligence when you want it.Frontier AI, enterprise by design.The AI workstation you actually own.Live web, tools, models and agents.Connected WEB. Isolated ZEG.
02
One platform, two boundaries.
Cortix has two distinct deployment boundaries. CortiX AI WEB is internet-connected by design. CortiX AI ZEG is the separate zero-egress enterprise server product.
CortiX AI ZEG
Zero-Egress Localized
Air-gapped enterprise deployment. Nothing leaves the machine. Built for regulated work where the network is the risk.
+Fully local inference, storage and memory
+Egress forbidden by a single fail-safe gate
+Local corpus search in place of the open web
+Covered by patent USA-P001
CortiX AI WEB
Public Hosted
A connected AI workstation designed to use live web research, remote APIs, model providers, MCP servers, plugins, cloud services and public resources as needed.
+Live internet search and retrieval
+Hosted multi-user sessions, federated access
+Identical panels and workflows to ZEG
+External network access enabled by design
CortiX
Single machine
One workstation. A local model serving the full studio on your own GPU.
NeuriX
LAN cluster
Several machines on a private network sharing one inference fabric.
MatriX
Cloud scale
Elastic cloud compute for the hosted edition and large workloads.
Deployment Pathfinder
A local, browser-only planning aid. It does not transmit your selections. Choose the operating boundary, compute scale and workload sensitivity to see the closest Cortix deployment pattern.
Interactive · local only
CortiX · ZEG
Recommended pattern: single-machine zero-egress deployment. Keep inference, memory and retrieval on hardware you control.
03
Eleven products. One inference runtime.
Cortix ships one shared inference runtime across browser clients and on-premises NVIDIA infrastructure. The portfolio currently comprises one shipping platform, two invite-only Alpha/Beta applications and eight products in development.
Shared Cortix runtime across connected browser inference and zero-egress enterprise deployment.
01
Cortix WEB
Platform · Individuals and technical teams
Browser-native AI inference platform running entirely client-side on WebGPU. Seven model profiles, a 2,090-skill library, agentic runtime, local memory and RAG, deterministic visual rendering and offline-capable PWA operation.
+No server round trip for the primary inference path; user data stays on the device.
+Progressive model profiles with quantized weights and a deterministic browser scheduler.
+Visual construction engine supports GLB, OBJ, STL and USDZ export.
7
Model profiles
2,090
Skill library
RTX
GPU acceleration
ShippingWebGPU · PWACurrent: RTX GPUsConsidering: CUDA · TensorRT
02
Cortix ZEG
Platform · Regulated enterprise
Zero-egress AI inference deployed inside the customer perimeter on NVIDIA hardware. Inference, retrieval and memory remain on premises with hardware-enforced egress control, SSO, RBAC, audit trail and tenant isolation.
+Reference architectures scale from RTX PRO workstations through H100-class clusters.
+NVLink intra-server and NVIDIA Quantum InfiniBand 400 Gbps fabric for cluster mode.
+Local NVMe model storage and NVML-based GPU visibility are part of the enterprise architecture.
0
Required cloud egress
400
Gbps cluster fabric
H100
Class scale
DevelopingCUDA · RTX GPUsConsidering: NIM · TensorRT-LLMNVIDIA AI Enterprise
VERTICAL APPLICATIONS
Domain products built on the same model loader, profile router, deterministic scheduler and cache.
03
Cortix E2Bizz
Vertical application · Commerce across four markets
Multi-market commerce super-app spanning the USA, Pakistan, India and the GCC with localized payments, logistics and compliance plus AI-assisted listings, search and a personal shopping agent.
+Web platform is live at e2bizz.com with mobile apps in invite-only testing.
+Per-market payment, logistics and regulatory configuration is built into the architecture.
4
Markets
Web
Platform live
AI
Listings · search · agent
Alpha/BetaAlpha/BetaCurrent: RTX GPUs (development)Considering: NIM · NeMo Retriever
04
Cortix Architect
Vertical application · Architecture and engineering
AI-assisted AEC drawing and design platform for credentialed architects and engineers with GPU-accelerated 2D drafting, 3D modeling, AI drawing generation, code-compliance checks and construction documentation.
+AI generation of plan, section and elevation drawings with deterministic professional review.
+Zero-egress deployment option for firms that cannot send project data to external APIs.
+Interchange workflows include IFC and DXF with Three.js and WebGL2 visualization.
2D
Drafting
3D
Modeling
AEC
Professional review
DevelopingCurrent: CUDA · RTX GPUsConsidering: Omniverse · RTX rendering · NIM
05
Cortix Tutor
Vertical application · Schools and training programs
Adaptive AI tutoring that diagnoses knowledge gaps, tracks per-concept mastery and changes explanation depth, hint sequencing and practice difficulty for each learner.
+Curriculum grounding uses institution-supplied materials through local retrieval.
+Deploys through browser-side WebGPU or school-controlled NVIDIA GPU servers with the same runtime.
+Teacher dashboard reports per-concept mastery across a class.
1:1
Adaptive tutoring
Local
Curriculum RAG
2
Deployment modes
DevelopingFERPA · COPPA alignedCurrent: RTX GPUsConsidering: NeMo · NIM · TensorRT-LLM
06
Cortix Game Suite
Vertical application · Game studios and developers
Node-graph game development environment with six GPU-accelerated generation modules spanning code, assets, NPC behavior, shaders, narrative and role orchestration.
+Visual node-based pipeline composition for end-to-end game creation workflows.
+Backing role infrastructure covers 476 roles, 731 skills and 1,723 role-skill mappings.
+Local or studio-controlled inference on the shared Cortix runtime.
AI legal research and drafting for law firms on the Cortix ZEG stack so privileged communications and case strategy remain inside the firm perimeter.
+Local legal research, drafting and citation workflows on on-premises NVIDIA GPUs.
+Matter-level isolation and audit trail designed for privilege protection and conflicts screening.
+Designed around confidentiality duties highlighted by ABA Formal Opinion 512.
ABA
Opinion 512
ZEG
Deployment
Local
RAG pipeline
DevelopingCurrent: CUDA · RTX GPUsConsidering: NIM · NeMo · H100
08
Cortix Ops
Vertical application · Infrastructure and IT operations
Self-healing AI operations control plane that performs live LLM root-cause diagnosis and executes supervised remediation through a validated autonomy ladder.
+Evidence-first repair flow: structured diagnosis, minimal patch and targeted then full validation.
+Automatic re-diagnosis after repeated failure while risky changes retain a human approval gate.
+Autonomy ladder has been validated through level 3 against a local model endpoint.
L3
Validated autonomy
LLM
Root-cause diagnosis
Human
Risk approval
DevelopingCurrent: CUDAConsidering: NIM · Triton Inference Server
09
Cortix Sigma
Vertical application · Analysts and researchers
Browser-based AI analytics extension for on-page data extraction, summarization and analysis with local inference, turning a web page into a structured data source without sending page content to external services.
+Chrome extension performs extraction and analysis directly in the browser.
+Local inference preserves page-content custody for restricted data-handling environments.
+GPU acceleration is used where WebGPU is available.
AI analytics engine with persistent charting, live data pipelines and a strict no-fake-data guarantee: every rendered metric traces to a real backend value or displays an explicit empty state.
+Global chart store creates chart instances once and updates them in place for stable dashboards.
+Evidence, forecast, graph and self-learning views operate over real data pipelines.
+Rendering policy rejects placeholder metrics rather than presenting invented analytics.
Vertical application · Trading and real estate investment
Round-the-clock autonomous trading and real estate investment platform with an institutional analytical toolkit where every metric traces to real market data.
+Reasoning and execution are separated: the reasoning tier never holds broker credentials or places orders directly.
+Every proposed position passes through a deterministic risk engine enforcing exposure and drawdown mandates.
+Portfolio, factor, drawdown and performance attribution use real executed fills with no placeholder or demo data.
24/7
Operating model
Risk
Deterministic gate
0
Placeholder analytics
DevelopingCurrent: CUDA · RTX GPUsConsidering: NIM · RAPIDS · Triton · TensorRT-LLM
04
One stack under every product.
Every Cortix product runs on the same shared core, so hardening done once protects the whole line. The egress gate sits at the boundary of the stack, not inside any single app.
Assembly with fixed precedence
System rules, roles, skills, memory and retrieval assemble into every request in a fixed order with token budgeting, so behavior is predictable and auditable.
Curated memory
Cross-session memory with importance scoring, near-duplicate reaffirmation, conflict surfacing, secret redaction on write and operator-controlled edit, delete and prune.
Verified search
The retrieval subsystem fuses sparse and dense ranking then runs claim-level ground checks before an answer ships, in both editions.
05
Connected WEB. Isolated ZEG.
CortiX AI WEB is the connected product: live internet research, APIs, MCP, plugins, remote models and authorized services are part of its operating model. CortiX AI ZEG is a separate enterprise server application where zero-egress is enforced for organizations that require isolation.
WEBConnected internet and service access
ZEGZero-egress enterprise server boundary
2Editions held at full parity every build
9Patent filings in the USPTO portfolio
ZEG only · enterprise server · zero egress
The zero-egress boundary
Every network call passes through one gate. In the ZEG edition that gate ships closed and fails closed, so inference, memory and retrieval resolve locally even under fault conditions. No telemetry, no callbacks, no silent updates. Data governance stops being a policy question and becomes a hardware fact. This boundary is the subject of patent USA-P001.
Two products · distinct network boundaries
Parity, not a stripped-down clone
CortiX AI WEB and CortiX AI ZEG share platform concepts but have intentionally different network behavior. WEB is connected-first and should never inherit a zero-egress restriction. ZEG is the enterprise server application where the network boundary is deliberately closed.
Connected by design
CortiX AI WEB may combine local inference with live web, remote models, APIs, MCP, plugins and public resources in one governed assistant runtime.
ZEG is separate
Zero-egress and fail-closed networking belong to CortiX AI ZEG and do not constrain CortiX AI WEB.
Explicit boundary
Deployment policy is a product choice: connected WEB for maximum reach or enterprise ZEG for isolated workloads.
Filed and claimedUnited States Patent and Trademark Office
USA-P001
Zero-egress localized enterprise AI
The boundary itself
USA-P002
mDNS-based auto-provisioning for local AI infrastructure
Fleet rollout without the cloud
USA-P006
Federated multi-user AI sessions
Teams on enterprise hardware
Filed May 11, 2026Six nonprovisional utility applications, now in USPTO formalities
Nine-patent portfolioHeld by United Services Associates Inc.
06
Affiliated ventures.
Alongside the product line, the United Services Associates group operates ventures that both use and prove the Cortix stack in live commercial settings.
Fund management
VanTerra Capital Holdings LLC
A Delaware management company serving as general partner and managing member of two private investment vehicles, with the VanTerra AI Trading System as its analytical engine.
+VanTerra Capital Equity Holdings Fund LP: long-only large-cap equity with an AI startup sleeve
+VanTerra Capital Holding Fund LLC: real estate across New York, New Jersey and Connecticut
+Institutional documentation and compliance program in place
vanterra.capital · Brooklyn, New York
Design and visualization
Atelier Vanterra Studio
A design coordination and visualization studio producing 3D architectural renderings, drawings, BIM and video for owners, architects and builders.
+Photoreal 3D rendering, animation and VR walkthroughs
+Drawing coordination and permit support for institutional projects
+Portfolio spanning college, school and civic buildings in New York State
Cortix Inc. is the AI systems company behind Cortix AI Studio and its affiliated product line. It operates as a subsidiary of United Services Associates Inc., a New York technology holding company incorporated in October 2012, holding a nine-patent USPTO portfolio that spans local inference, auto-provisioning, federated sessions and secure AI operations.
Cortix Inc.subsidiary of United Services Associates Inc.New York, USA
Registered office
Cortix Inc. 138 Ormonde Blvd Valley Stream, NY 11580 United States
Parent company
United Services Associates Inc. 646 Coney Island Ave Brooklyn, NY 11218 Incorporated October 2012
Hyper-real tactical universe · two simulations · zero egress
KeyboardArrows or WASD move · Space fires, hold for rapid · Shift raises shield · P pauses · F fullscreen
TouchOn-screen pads move, fire and shield
GamepadLeft stick moves · A fires · B shield · right trigger thrusts
StarshipSwarms attack in formations of 3, 7, 9, 13, 17, 21 and 27. Kills earn credits. Spend them at the starbase between waves on weapons MK1 to MK5, rapid overdrive, shields and hulls. Elite difficulty pays 1.5x credits.
Top pilots · this device
Paused
Starbase dock
Field dock · refit, then return to the fight
Credits 0
Perimeter held?
Score 0 · Best 0
Top pilots · this device
Post your run to the board. Name, email and phone go to the Cortix team at info@cortixaistudio.com so we can get in touch.
Request access
Tell us about your team and what you want to run. Your message is sent securely from this form directly to our inbox. No email application is required.
info@cortixaistudio.com
Runtime ManagerStage 11
0Workers
0 MBModel memory
-Safe budget
Model lifecyclelocal
Model memory values are profile estimates. Browsers do not expose reliable dedicated GPU VRAM totals, so the safety budget uses a conservative browser memory hint.
Retrieval hardeningRRF + rerank
Stage 11 retrieval: query variants, reciprocal-rank fusion, WebLLM reranking, duplicate suppression, 90-second retrieval cache and page/slide/sheet provenance when available.
