Sentinel: SEO, GEO and AEO observability, self-hosted.
Sentinel is a self-hosted SEO, GEO and AEO observability and opportunity-engine platform. It watches what humans search for, what LLMs say about a brand, what gaps exist between site claims and model beliefs, and what the corpus a business doesn't control is telling the world, then produces a ranked queue of concrete actions.
The source is private and Sentinel is internal to Willhite Strategy Group; there is no public repo or public dashboard to link.
Repo stats
Commit history and language mix, live hourly. The repo is private, so there is no star count.
Why I built it
The working thesis is what I call the Field Theory of Machine Confidence: the next moat in branding is not visibility, it is verifiability, surviving an audit conducted by an LLM acting as a fiduciary for the user. Sentinel is built around three consequences of that idea. A brand that says nothing about a topic doesn't read as "no known issues" to a model, it reads as a blank the model fills in from priors, so silence gets filled. Most of what a model believes about a brand is assembled from sources that brand doesn't own, so the corpus is the lever, not the campaign. And a multi-agent buyer convenes several lenses at once, engineering, legal, finance, security, marketing, and one weak lens can sink the whole verdict.
Sentinel and its companion system, Strider, are kept as two separate things on purpose: Sentinel observes, measures and reports, and Strider acts on what Sentinel finds. Anything that reads, measures, scores or renders is Sentinel; anything that writes, commits, deploys or closes is Strider. Keeping the two apart means Sentinel's report can be sold on its own, to a client who never has to let an agent near their site.
What it does
Three layers, from observe to operate.
- Monitor. Site health checks, Google Search Console performance, AI-citation tracking across five LLM engines, rank tracking, bot crawl logs, PageSpeed Insights, schema audit, crawl audit, brand-mention monitoring, Bing Webmaster data, backlink authority, and a directory-presence check (Google Business Profile, Yelp, BBB and category-specific listings).
- Analyze. Multi-engine AEO grading, a five-lens swarm cross-examination, silence-gap detection (site-stated facts vs. LLM-stated facts, side by side), source attribution across the third-party domains a model actually cites, and anomaly detection on search performance.
- Act. A ranked queue of content briefs, scored from search-trend, competitor and AEO-gap signals, plus a corpus-shaping queue naming exactly which outside source to go influence next.
Stack and architecture
Python workers on a schedule, each one a small, focused job that reads or writes one slice of the picture, backed by Postgres. A Flask dashboard renders the combined view per property: health, search performance, AI citations, briefs, bot crawl, AEO grades, swarm verdicts, belief sources and silence gaps in one pane. A budget guard enforces a hard monthly spend cap per client before any paid LLM call runs, so a runaway job cannot produce a surprise bill.
Status
Live and running nightly against Willhite Strategy Group's own site portfolio and its active web-design clients. Internal only: there is no public dashboard or self-serve signup, and it is never pointed at a prospect on an ongoing basis.