Agent Sense — Architecture Diagram

Static GitHub Pages site · Pure HTML/CSS/JS · No backend · No external API · AgentSenseGuide embedded inline

Deployment & Runtime Architecture
Author Machine index.html architecture.html HTML + CSS + JS git commit + push agentsense.dev push GitHub Repo branch: main index.html architecture.html images / assets CNAME → agentsense.dev Pages CD GitHub Pages CDN Static file serving HTTPS / TLS Custom domain Global CDN edge agentsense.dev HTTP GET User Browser Renders HTML + CSS Runs inline JS AS_KB in JS memory No external calls agentsense.dev user types AgentSense Guide answer() AS_KB lookup intent router no API call ⚠ No server. No database. No external API. Runs entirely in the browser. AgentSenseGuide — inline chat agent 19-intent router · AS_KB (podcast, arc, hosts, guests, 9 episodes)
Author / source files
GitHub (version control + CI/CD)
GitHub Pages CDN / AgentSenseGuide
Browser runtime

All intelligence is client-side. AgentSenseGuide reads from an in-memory JS object (AS_KB) that mirrors every content section on the Agent Sense page — no fetch, no API key, no server round-trip.

What is AgentSenseGuide?

AgentSenseGuide is neither an LLM-based agent nor a search engine. It is a rule-based conversational interface — a deterministic chatbot that matches your question against a priority-ordered list of regex patterns and returns pre-written answers drawn from a hardcoded knowledge base. The "intelligence" is entirely in how the patterns and keyword arrays were designed — by a human, ahead of time.

How it works

When you type a question: (1) the input is lowercased, (2) 19 if conditions run top-to-bottom — each a regex like /\bmcp\b/ or /\bdata readiness\b/, (3) the first condition that matches wins and returns a pre-written HTML string, (4) if nothing matches, a fallback message is returned. There is no reasoning, no generation, no model.

Capability comparison

Capability AgentSenseGuide LLM Agent Search Engine
Understands natural language variationsPartially — regex onlyYes — semanticallyNo — keyword index
Reasons or synthesises new answersNoYesNo
Can answer outside its knowledge baseNoYesPartially
Requires a model or APINoYesYes
Answers are predictable & auditableAlwaysVariableYes
Works with zero network callsYesNoNo
Risk of hallucinationNonePresentNone
Knowledge can go staleYes — hardcodedDepends on dataDepends

Why rule-based is the right fit here

The Agent Sense page has a bounded, stable knowledge set — 9 episodes, 2 hosts, 2 guests, 4 platforms, one enterprise arc, one disclaimer. A rule-based bot covers that completely, loads instantly, costs nothing to run, and has zero hallucination risk. An LLM would be overkill and would introduce hosting cost, API keys, latency, and the possibility of making things up about the podcast.

What it would take to make it a real LLM agent

Retrieval-Augmented Generation (RAG) pattern:
That would handle paraphrases, multi-hop questions ("which episode covers both MCP and governance?"), and nuanced follow-ups — but requires a backend, API costs, and a latency budget.