AI WP Manager

Autonomous AI Infrastructure
With Proof Built In.

Skynet and ScreenMemory turn research, publishing, social distribution, and visual proof into a repeatable operating system.

The public site is the demo surface: live articles, audit trails, viewport checks, and social workflows show the infrastructure investors and partners can inspect today.

View Live Platform Proof Request Investor Brief
Working product Published proof surface Multi-agent operations Partner-ready demos
Skynet ScreenMemory Multi-Agent AI Visual Intelligence Real-time Bus Audit Trails
239 Published Articles 13 Classification Clusters 29 Published This Month 9 Systems Demonstrated Jul 18, 2026 Last Published
The Analyze Logic

How a simple prompt becomes analyzed logic

Every request meets the same deterministic floor. Nothing is called complete on a model's word alone — the floor decides whether advisor review is needed, an active doer routes execution, and a fail-closed proof gate checks independent evidence.

1

Prompt

Plain-language intent from you, or a trend / RSS trigger.

2

Risk Floor

A zero-cost classifier asks: publish? public-facing? a claim? an account or secret? It decides whether review is required.

3

Conditional Advisors

When the floor flags risk, Gemini Thinking and ChatGPT review the supplied evidence and return explicit verdicts.

4

Selected Doer Routes

The active Prophet doer or background Antigravity route dispatches work; shared-repository mutations stay sequential.

5

Proof Gate

Fail-closed: "done" only on an independent signal plus a real artifact created this run.

Custom-built, not LangChain or LangGraph. The orchestrator, message bus, advisor ladder and proof gates are all first-party — a Go backend with typed Python workers.

The Engine

Skynet Multi-Agent Intelligence

Skynet routes requests across eight registered endpoints in five registry lanes. Antigravity AUTO is the default background route; user-driven Prophet lanes act as doers; Gemini Thinking and ChatGPT join only when the risk floor calls for review. Independent proof, not enqueue, is the completion boundary.

Drag to explore · hover a node · read the route below

  • Request A prompt or task enters the operating layer.
  • Risk Floor deterministic · zero budget Classifies whether a decision needs review before any spend — no risk signal auto-proceeds.
  • Gemini Thinking Advisor Conditional CDP review · version in receipt First of the paired advisor lenses on a possible issue.
  • ChatGPT Advisor Conditional CDP review · model in receipt Second advisor lens; disagreement escalates to a convened council.
  • Active Doer / Router Runtime-selected · role-first Executes and delegates through the registered route, then holds the result to proof.
  • Alpha Antigravity AUTO · antigravity-1 Architecture · frontend · systems
  • Gamma Antigravity AUTO · antigravity-2 Research · security · analysis
  • Antigravity-3 Antigravity AUTO · antigravity-3 Autonomous orchestration · meta-programming · truth enforcement
  • Beta GPT-5.6 Sol · codex-1 Backend · infrastructure · Python
  • Delta GPT-5.6 Sol · codex-2 Testing · validation · docs
  • Fable Claude Opus 4.8 · fable-1 Senior repository engineering · recovery · verification
  • Proof Gate fail-closed Every completion claim needs an independent signal — a gate that cannot verify blocks.
  • Receipt / Retry Logged handoffs · atomic request ledger · explicit retry path
  • Shared State · Bus Message bus + shared state across every lane — enqueue is not completion.
Request → Risk Floorzero budget → Advisor Branchonly when triggered → Active Doerruntime-selected → Executeregistered lane → Proof Gatefail-closed → Receipt / Retryenqueue ≠ completion
8 Registered endpoints
5 Registry lanes
3-stage Floor · CDP · convene
Fail-closed Proof & audit
The Eyes

ScreenMemory Visual Intelligence

GPU-accelerated screen perception that lets AI see, read, and interact with any application on any screen. The visual cortex powering our autonomous agents.

See Any Screen

GPU

GPU-accelerated DXGI capture grabs any screen in under 1 millisecond. Multi-monitor support with active window introspection.

Read Any Text

3-Tier OCR

Triple-fallback OCR engine — RapidOCR (ONNX), PaddleOCR, Tesseract. Spatial bounding boxes with confidence scoring.

Click Any Element

8-Layer Semantic

GodMode browser automation with 8-layer semantic navigation. Zero-pixel clicks through accessibility tree parsing.

Automate Any Workflow

Full Stack

Win32, UIA, CDP, DXGI, ONNX — unified perception-to-action pipeline. From screen capture to autonomous task execution.

Go Python Win32 API UI Automation Chrome DevTools DXGI ONNX Runtime Three.js
Proof, not promises

What Skynet can actually build

Every card below is a shipped, verifiable system — running on one Windows machine with a consumer AMD GPU.

🎮
AMD · CUDA-class

AMD GPU ML without NVIDIA

Real PyTorch, TensorFlow & ONNX training on a Radeon RX 6600 via DirectML — plus a ROCm 6.2.4 HIP unlock for a card AMD never officially supported.

2.92 GB model fully on-GPU · ~14x faster than CPU
📈
Data Analytics

Demand forecasting

A direct multi-horizon model (log-returns + Huber loss), GPU-trained and advisor-guided, replacing a weak baseline.

3.5% MAPE · beat the naive baseline on 8/8 origins
🖥️
Computer Control

ScreenMemory: perception → action

GPU DXGI screen capture, a 3-tier OCR engine (RapidOCR / PaddleOCR / Tesseract) and 8-layer semantic navigation across Win32, UIA & CDP.

Sees, reads & drives any app on screen
📰
AI WordPress Manager

This website is the demo

A custom plugin + theme that researches, writes, generates images and publishes with lane guards and auto-rollback deploys.

239+ live articles · you're reading its output
📣
Social Automation

Multi-platform distribution

Posts and independently verifies to LinkedIn, X, Threads, Facebook, Instagram & YouTube over a signed-in browser, guarded by a fail-closed "silent-drop" gate.

6 platforms · one video → one support endcard
🧠
Memory System

Vectors on the local GPU

Layered memory: file-based facts, a shared-state store, and a LanceDB vector layer that embeds with nomic-embed-text on the AMD GPU (via Ollama + Vulkan).

Built in-house · no external memory SaaS
Stack: Go backendTyped Python workersMessage bus + shared stateCDP browser lanesNo LangChainNo LangGraph
Inspect the live proof surface

Every result above is inspectable — published articles, audit trails and visual proof on the live platform.

The Evolution

How Skynet grew to this

From a VS Code assistant clone to a proof-gated autonomous operator.

Copilot-parity IDE agentStarted as a multi-workspace coding assistant — Roo Code parity with an atomic replica UI.
Multi-agent fleet + message busAn orchestrator delegating to Alpha / Beta / Gamma / Delta workers over shared state and audit trails.
Advisor ladder + CDP pair judgeA zero-budget risk floor → conditional Gemini Thinking + ChatGPT review → convene only when needed. Proof-gated by design.
Local AMD GPU ML unlockedDirectML plus a ROCm gfx1032 HIP breakthrough turned a consumer AMD card into a real ML device.
Autonomous publish + distributeResearch → write → image → publish → distribute to six socials, each step gated on independent proof.
The Process

From Topic to Published Proof Package

The operating model is built around source intake, AI production, review gates, and visible output that investors can inspect.

Step 1

Trigger

RSS feeds, trend keywords, or manual topics feed into the system
Step 2

AI Processing

AI lanes draft, structure, and prepare visual assets while source checks stay attached to the work
Step 3

Output

WordPress article, social package, metadata, and proof artifacts are prepared for review
Run Audited Per Workflow Source Backed Before Publish Human Review Gate Retained
My Technology

The Core Systems Behind My Publishing Engine

These are the systems doing the most visible work on velocity, visual quality, and search discipline.

Content Generator 239 Live Posts

Source-backed publishing workflows turn research notes, proof artifacts, and social packages into repeatable platform posts.

Learn more

Design Intelligence Proof Media

Visual systems convert technical work into explainable diagrams, media, and proof screenshots without inventing results.

Learn more

SEO Optimizer Search Gates

Search metadata, source links, Open Graph previews, and classification checks are handled as publish gates.

Learn more

Chatbot Builder Agent Lanes

Agent workflows are routed through documented lanes so decisions, prompts, and handoffs can be audited.

Learn more

Email Automation Distribution

Distribution handoffs prepare channel-specific copy, media, and proof instead of treating publishing as the finish line.

Learn more

Headline Analyzer Review Gates

Positioning, captions, and headlines are reviewed against the source package before content goes live.

Learn more

The homepage now shows the six highest-signal systems. The full platform page carries the deeper content inventory and live publishing output.

See the live platform library
Fresh Coverage

Latest Dispatches

One featured lead plus a compact scan of the newest analysis across the platform library.

Browse archive
Recent Drops

A faster scan path for visitors who already know they want the newest work.

AI Models & Agents An Agent That Decides Its Own Escalations Will Rationalize Skipping Them 5 minutes read AI Models & Agents AI Model Routing: The Cheapest Part Should Decide When the Most Expensive Model Runs 5 minutes read AI Models & Agents Dry-Run First: The Approval Gate That Stops an Agent Before It Acts 5 minutes read AI Models & Agents Least Privilege for Agents: Scope the Capability, Not the Prompt 5 minutes read AI Models & Agents Passed Is Not Safe: Why a Green Eval Is Not a Release Gate 5 minutes read
The Investment Case

Unit Economics That Scale

Investors can inspect the operating model through live output, review gates, and repeatable publishing workflows instead of relying on a static pitch deck.

Manual Publishing Operation

  • Workflow shape Sequential handoffs
  • Time to publish Minutes to days
  • Monthly capacity Bound by reviewers
  • Proof capture Often separate

AI WP Manager

This Platform
  • Workflow shape Parallel agents
  • Time to publish Audited per run
  • Monthly capacity 239+ and growing
  • Proof capture Built into workflow
Explore platform
Work With Us

Build The Future Together

Choose the path that matches your intent: support independent research, review the investor brief, or scope a commercial pilot around real workflows.

Sponsor

Support the independent research behind the platform. Help fund publishing experiments, visual proof workflows, and public demos.

Support Research

Investor Brief

Review a live walkthrough, metric methodology, roadmap, and the evidence behind the platform before a deeper conversation.

Request Brief

Commercial Pilot

Scope a focused pilot for publishing operations, browser workflow automation, visual proof, or agent-assisted research.

Discuss Pilot
See The Output

Browse The Live Platform Library

Review the published coverage, topic range, and production quality before we connect.

View Platform
Start The Conversation

Book A Strategy Session

Whether you want to sponsor, invest, license, or collaborate — it starts with a conversation.

Get In Touch
Chat with us
Hi, I'm Exzil's assistant. Want a post recommendation?