Consolidating Super Search, OSINT Person, and ClearFront into one intelligence gateway, then adding the one leg none of them has: watch.
We have already built most of an intelligence platform and left it scattered across three surfaces. Three MCP servers run live today: Super Search (26 tools), OSINT Person (17 tools), and ClearFront (31 tools). That is 74 tools running as three separate surfaces, and the third one, ClearFront, is completely unproductized.
What is missing is not another search tool. It is the fourth leg: watch. Today we pull on demand (Super Search) and profile one entity deep (OSINT Person). We do not watch anything. The monitoring we do run is scattered across cron jobs and skills.
This memo lays out the build: unify the three surfaces into one gateway, then build a monitoring and alerting engine on top of it. One engine, five products feed from it.
Multi-provider search across SearXNG, Exa, Brave, Serper, Tavily, Perplexity, Firecrawl, CourtListener, and the Wayback Machine. Fused search, racing search, deep research, academic, news, images, document extraction, and screenshot capture.
Deep, one entity. Person search, phone and email lookup, property records, business affiliation and entity search, court records, data-broker extraction, dossier rendering, and result reconciliation.
Username sweep across 3,400+ sites, email breach (Hudson Rock, HIBP, EmailRep), domain and IP recon (crt.sh, Shodan, Censys, GreyNoise, Wayback, WHOIS, DNS, theHarvester), EXIF and GPS, paste, crypto, and dorks. Audited August 5, 2026 as ADOPT. MIT, keyless for most sources. Fully deployed, completely unproductized.
Today we have pull and profile. We do not have watch.
The monitoring we do run today is scattered: blog watcher, competitor news, and a recurring scout, each its own cron job and skill. The build consolidates them into one service.
| Product | What watch does for it |
|---|---|
| IntelSight | Competitor monitoring is its whole pitch, done ad hoc today |
| OSINT Person | Standing person watchlists instead of one-shot skip traces |
| VerdictTank / RFP Tank | New-RFP and solicitation alerts |
| DRE | Debtor movement and new-asset detection |
| NVR Shield | Already ships bespoke monitoring, proving the pattern |
The service is one loop: standing query → schedule → diff detection → alert → digest.
Super Search, OSINT Person, and ClearFront are three separate surfaces with overlapping tools. Username and email search already exist in both OSINT Person and ClearFront. Merge them into one MCP/API surface with a shared provider router, a shared cost layer, and one billing surface.
Mostly wiring, not new code. The clearfront audit already anticipated this: "wrap Hudson Rock + Maigret as providers in OSINT Person."
With one surface underneath, the monitoring engine writes once and reads everything. Standing queries, schedules, diff detection, alerts, and digests, all against a single gateway.
One MCP/API surface, shared provider router, shared cost layer, one billing. Collapses the overlap and makes everything downstream cheaper.
Search once, remember. Results resolve against a graph instead of being throwaway. This is what turns search into intelligence.
The obstacle-as-product move. Multi-provider search with circuit breaking and failover is a product other builders pay for. IntelSight is already proof it can carry a customer.
Show which provider answered, cost per query, and fallback health. That is the differentiator against a bare search API.
| Candidate | Why we set it aside |
|---|---|
| Shared SaaS scaffold | High leverage but plumbing; Stack Auth already covers much of it |
| Entity knowledge graph | Powerful but internal-first; better as a layer under the tools than a standalone product |
| Productizing Crawl4AI alone | Too narrow; it is a component, not a product |
Watch needs the unified surface underneath it. Build watch first and you wire it three times, once per surface. Unify first and you build watch once, on one surface.
Sequencing matters because the second phase is cheap only if the first is done.