Market intelligence · Early access

Give your AI a live intelligence layer for the smart home market.

Connect Claude, ChatGPT or your own tooling to a structured, continuously updated read on what is actually happening in smart home: products, platforms, integrations, launches, compatibility claims, complaints, and the feature requests buyers keep repeating.

Your AI already knows how to reason. We give it something worth reasoning about.

What's in it today

117,000+
structured mentions entities resolved, typed, dated
14,400+
sources behind them articles, reviews, forum threads

Every mention typed by what it says:

  • compatibility
  • issue
  • review
  • deal
  • launch
  • comparison
  • user need

With a signal layer on top - launches, vulnerabilities, sentiment shifts, partnerships, protocol milestones - each one traceable back to the mentions it was built from.

your assistant, connected to the corpus

Example question: We position on three things. Which of them do buyers repeat back? Answer: All three are in the corpus and they land very differently. Claim one is echoed 110 times, and it is simultaneously your largest complaint cluster: 21 of the 59 logged issues. Claim two lands warmly but rarely, around 12 mentions, with a competitor credited alongside you for it. Claim three appears 7 times, and six of those say you do not have it.

Scripted session. Every number in it is a real, queryable value from the corpus.

Why a corpus and not a search

Your assistant is smart. It has nothing to reason over.

A web search returns pages, and a page is where the work starts. Here the work is already done: every mention is extracted from the source, resolved against a curated vocabulary of brands, platforms, protocols, device types and products, and typed by what it actually says.

Typed, not just found

A mention is not a page that matched a keyword. It is a compatibility claim, an issue, a review, a deal, a launch, a comparison or a stated user need, carrying its sentiment and a verbatim quote from the source it came from.

Signals on top of mentions

Launches, vulnerabilities, sentiment shifts, partnerships and protocol milestones are synthesised from the mentions underneath. Your AI can ask what changed this month instead of reading everything published in it.

Evidence all the way down

Every aggregate drills back to the exact sentence it came from. That is also how you check one: slice the community threads out, isolate the editorial reviews, and see whether a finding survives its own sample.

Coverage is deeper in some categories than others. Four of them today:

4,751
smart lock mentions across 970 articles
4,510
robot vacuum mentions across 813 articles
1,661
robot mower mentions across 309 articles
1,322
smart blinds mentions across 411 articles

Real output, real data

Questions a web search cannot answer

Findings an assistant produced with nothing but a connection to the corpus. Every number is queryable and every claim traces back to sources you can open and read.

Product · smart locks

What are the deal breakers in our category?

The largest issue cluster is locks physically failing to lock or unlock, with firmware updates as a repeated trigger. Second, battery life collapsing with no spare available to buy. Third, headline features like hands-free unlock failing across vendors rather than at one of them.

The corpus also flags its own bias: one brand dominates the raw counts because its community forum is heavily ingested. Re-sliced to editorial sources only, the brand spread flattens, and all three themes still hold.

base: 4,751 mentions · 970 articles · sentiment 1,265 positive / 1,075 negative

Partnerships · compatibility

Audit our "works with everything" claim.

For one switch and relay maker, press coverage was compared against user reports platform by platform. Compatibility is asserted at almost the same rate in both. Problems are reported more than four times as often by the people who actually installed the devices.

The verdict per platform: one ecosystem genuinely strong, one a documented liability where the working path is a volunteer’s driver, and two where 33 press claims stand against zero user evidence in either direction.

base: 1,053 mentions · 374 press · 679 community

Strategy · smart blinds

If you were us, what would you do differently?

For one blinds maker, the most damaging user verdict about its product lives on a competitor’s review page. Buyers also discuss blinds in a Home Assistant context nearly three times as often as in the Apple context its marketing targets.

Output: five prioritised moves, each carrying the mention count behind it, from closing a known-defect story to leading with "no cloud", which appears as a stated purchase requirement in multiple threads.

base: 1,322 mentions · 411 articles · 6 brands ranked by volume

Who it is for

One corpus, three teams

Product

Roadmap with evidence

Which stated needs in our category does nobody ship yet, and how many buyers voiced them?

Marketing

Speak the buyer's language

Which words do buyers use for this problem before they know we exist, and which competitor claims stick in reviews?

Partnerships

Know your real position

Where is our integration story genuinely strong, where is it a liability, and where do we have claims but no evidence?

How you connect

One server. The tools your team already uses.

No dashboard to learn, no query language, nothing for your team to install beyond a setting they paste once.

An MCP server

Add one remote server to Claude, ChatGPT or any MCP-capable tool. Your team then asks in plain language, and the assistant queries the corpus itself: which mentions, of what type, about which brand, from which kind of source, over which period.

It comes back with an answer and the evidence under it, in the tool they were already working in.

Something other than MCP?

If the intelligence needs to reach a dashboard, an agent of your own or a spreadsheet instead, say so when you get in touch. Early access is where it gets decided what gets built next, and there is no point building the second door before someone needs to walk through it.

Early access

The intelligence layer is ours. The questions are yours.

We are opening this to a small number of teams first, so that each one gets set up properly and so that what they ask for shapes what gets built next.

Tell us your category and we will tell you what the corpus already holds on it, before you commit to anything.

No credit card, no sales sequence. One reply from a person, usually within a day.

Want to be kept in the loop?

Leave an address and you will hear when access opens, and occasionally when something worth knowing lands: new categories, new signal types, what early teams are asking it. Not the newsletter, unless you subscribe to that separately.