Technical Deep Dive

Architecture
of a Learning Browser

12 core modules in 5 categories. Each system explained in its own video — pick a category and dive in.

The central learning loop — four systems, one cycle.
LCJ
Observation
Candidate Journal
ALP
Quality
Learning Pipeline
PKS
Memory
Knowledge Store
CLS
Verification
Closed-Loop
150+
MCP Tools
55+
SQLite Tables
25
Learning Modules
140k+
Lines of C#

Core Modules in Detail

Pick a category — each module has its own explainer video.

PKS
Knowledge Store
Phenomenological Knowledge Store
Long-term memory for website patterns. Each domain is stored as a collection of phenomena — with multi-fingerprints across 5 dimensions, reusable playbooks and continuous health tracking.
Watch explainer video
LCJ
Candidate Journal
Lightweight Candidate Journal
The real-time observation log. Every agent interaction is captured before anything is filtered — activity sessions, tool events, candidates and observations form the raw data foundation.
Watch explainer video
ALP
Learning Pipeline
Agent Learning Pipeline
The bridge between raw data and verified knowledge. Observations are evaluated through formal quality gates and automatically promoted through three maturity levels: L0 → L1 → L2.
Watch explainer video
OK
Operational Knowledge
Operational Knowledge System
The real-time knowledge base: Which service is open, which account is active, which capabilities are available? Append-only observations are aggregated into versioned facts.
Watch explainer video
CLS
Closed-Loop
Closed-Loop System
The verification cycle. Every agent action is checked: precondition check, dispatch, outcome verification, learning feedback. Result: satisfied, failed or unknown.
Watch explainer video
EVM
Evidence Verification
Evidence Verification Mode
Fact-based research with formal source requirements. Every claim is decomposed into testable statements. Critical domains require at least 2 independent sources.
Watch explainer video
ETM
Task Memory
Episodic Task Memory
Recognizes recurring task types and improves with each execution. Task profiles store guidance and mandatory checks. Prevents premature termination of large tasks.
Watch explainer video
Crawl
Embedded Crawler
BFS Website Crawler
A background crawler that explores website structures via breadth-first search. Extracts metadata, detects frameworks, captures screenshots and performs hydration drift validation.
Watch explainer video
DUE
Deep UI Explorer
Surface Explorer
Discovers and safely activates hidden UI content in SPAs: modals, tabs, accordions, popovers — everything without its own URL. Safety-based classification.
Watch explainer video
AAG
Awareness Gates
Agent Awareness Gates
Guard system for correct agent initialization. Four gates: Bootstrap, Perceive-First, SPA Navigation and ETM Task Discovery. Warn first, then block.
Watch explainer video
ASD
Auth Detection
Auth Surface Detection
Universal authentication detection. Recognizes login forms, MFA prompts and auth errors on any website. Tri-state: Yes, No, Unknown.
Watch explainer video
MCP
MCP Server
Model Context Protocol — 150+ Tools
JSON-RPC 2.0 over localhost TCP, secured by bearer token auth. Any MCP-capable AI agent can use Nova — Claude, GPT, Gemini, or custom scripts.
Watch explainer video