What Archevi Actually Is (and Why We Built It)

Every family has a version of the same problem. The documents exist -- insurance policies, tax returns, medical records, property deeds, warranty cards. They're in a drawer, a folder on Google Drive, a shoebox, an email attachment from 2021. The information is there. It's just never findable at the moment you need it.

We didn't set out to build a document storage app. There are plenty of those. We set out to build something that could answer a question.

The Question That Started This

The trigger was specific. A family member needed to know whether a medical expense was covered under their health insurance plan. The policy document was somewhere -- probably in a "Documents" folder, possibly in an email, maybe still in the original envelope from the insurer. Finding it took 40 minutes. Reading the relevant clause took another 10. The answer was two sentences buried on page 14.

That's not a storage problem. Google Drive could store that PDF perfectly well. It's a retrieval problem -- knowing what's inside your documents without having to open and read each one.

What It Does

Archevi is a family knowledge base. You upload your documents -- PDFs, photos of receipts, scanned letters, whatever you have. From that point, you ask questions in plain language:

• "When does our home insurance expire?"
• "What's the deductible on our car policy?"
• "Which medical expenses can we claim this tax year?"

Archevi reads your documents, finds the relevant sections, and gives you an answer with the source highlighted. Not a keyword match -- a semantic understanding of what you're asking and what your documents contain.

The difference matters. If your municipal tax assessment is saved as "2025_city_notice.pdf" and you search for "property tax deadline," a traditional file search returns nothing. Archevi finds it because it understands what the document is about, not just what it's called.

How It Works (Without the Jargon)

When you upload a document, we extract the text and convert it into a mathematical representation that captures its meaning -- what AI researchers call an embedding. Your question gets converted the same way. Then we find the document sections whose meaning is closest to your question.

Those sections get passed to a language model that generates a clear answer, citing the specific part of your document it drew from. You can verify the source yourself -- we're not asking you to trust a black box.

The whole pipeline runs on infrastructure we control. We use Cohere for the embeddings, Groq for the fast inference, and PostgreSQL with pgvector for the search. No data gets stored on third-party servers beyond what's needed for the AI processing step -- and even that goes through a privacy layer first.

The Privacy Decision

Family documents contain sensitive information. Social Insurance Numbers, medical records, financial statements, your children's school records. We made an early decision that shaped everything else: your data should never reach a cloud AI service in identifiable form.

So we built an anonymisation layer. Before any document text is sent to an AI model, personally identifiable information -- names, SIN numbers, addresses, phone numbers -- is replaced with surrogates. The AI sees "PERSON_1 has a policy with INSURANCE_CO_1" instead of your actual name and insurer. The real information is restored in the final answer.

Is this perfect? No. The privacy layer adds latency -- responses take a beat longer than they would without it. Occasionally it over-redacts a common word. We decided the tradeoff was the right one: slower is acceptable; exposed is not.

Why Canada Matters

All document storage runs on a server in Toronto. Not because Canadian hosting is inherently more secure -- the bits don't know what country they're in. It matters because of jurisdiction.

Under the US CLOUD Act, American authorities can compel US companies to hand over data stored anywhere in the world. If your family's documents sit on AWS or Google Cloud, they're subject to that authority regardless of whether you're Canadian. By hosting on Canadian infrastructure under PIPEDA, your data stays under Canadian privacy law.

This isn't a marketing point. It's an architectural decision that constrains where we can build and what services we can use. We chose the constraint deliberately.

What It Costs

The Starter plan is $6/month. For that, a couple gets semantic search across all their uploaded documents, the privacy layer, Canadian hosting, and the ability to ask questions instead of searching file names.

We estimate the practical value at $300-700/year in avoided costs -- missed insurance renewals, unclaimed tax deductions, duplicate services you forgot you were paying for -- plus 15-20 hours of time not spent searching for documents. The math isn't dramatic. It's just consistently positive.

What It Doesn't Do

Archevi is a retrieval system, not an advisory one. When it tells you your insurance expires on March 15th, it's reading your policy document -- not recommending whether you should renew. When it finds a medical expense that may be claimable, it's surfacing what the CRA's guidelines say -- not replacing your accountant.

The value is narrow but real: getting the right information from your own documents at the moment you need it. For most families, that's the part that wasn't working.