Quill Reads Canada

Ask a chatbot a plain question about Canada, who lobbied which committee last spring, what a regulation actually changed, what a city published in its latest open dataset, and you will get an answer delivered with total confidence and no receipt. Sometimes it is right. Sometimes it is a very fluent guess wearing a suit. The model is not lying on purpose; it simply never read the record, because the record is not in one place. It is spread across federal portals, thirteen provincial and territorial legislatures, hundreds of regulators, city data portals, court databases, gazettes and lobbying registries, each with its own format, its own search box and its own idea of what a date looks like.

Quill¹ is my answer to that mess, and the tagline on the project is the whole pitch: Quill reads Canada so your AI doesn’t have to lie about it.

What It Actually Is

Quill is a data pipeline, the plumbing that goes out, fetches records, checks them and stores them, pointed at Canada’s public sector. Right now it registers 207 data modules, each one a small program that knows how to pull and parse one kind of source: Statistics Canada tables, House and Senate committee evidence, Hansard, the Canada Gazette, the federal lobbying registry, procurement notices, CanLII court decisions, provincial legislatures and open-data portals, municipal datasets and more. On top of that pipeline sits an MCP server, the Model Context Protocol, a standard plug that lets an AI assistant call outside tools, with 184 tools. Beside it sit a REST , the ordinary web interface that programs talk to, and client libraries for TypeScript and Python. The portal speaks English and French.

The point is not that Quill knows everything. The point is that when an agent asks it something, the answer comes back with where it came from.

Receipts, Not Vibes

Every record Quill ingests can carry a SHA-256 hash, a short fingerprint computed from the exact content, together with the source it came from, when it was retrieved and whatever licence the publisher declared. The high-signal answers carry a list of claims built the same way. If Quill publishes a hash, you can recompute it yourself with one line in a terminal and compare, without trusting Quill at all. Checking a claim is free, always.

And the project says plainly what a hash does not do, which is rarer than it should be. The provenance principle on the site reads: “A hash checks integrity, not authenticity or reuse rights. Source terms still apply.” A fingerprint proves the record has not changed since Quill saw it. It does not prove the publisher was right, and it does not hand you permission to republish someone’s dataset. Public availability is not a licence. I would rather ship a tool that tells you the limits of its own evidence than one that sells certainty it cannot back up, because certainty is the cheapest thing a model produces, and Zero-Trust already covered what I think of that ask to be believed.

Who It Is For

The journalist with a lead, the researcher with a question, the developer with an idea. That is the line on the front page, and it is accurate. A reporter can follow a name from the lobbying registry into committee appearances and Governor-in-Council appointments, because the engine underneath resolves the same person or company across registries and draws the connections. A researcher can pull a Statistics Canada series and its source link in one call. A developer can point their own agent at the MCP , or skip the hosted service and run the server themselves.

The hosted endpoint has a free tier of 1,000 calls a month, with paid plans for heavier use. calls are not counted against anything, because they never touch my servers.

Why Self-Hostable Matters

Data about Canada should not depend on whether a foreign platform keeps its pricing page stable. Quill runs on hardware I own, with no cloud bill, behind a Cloudflare tunnel, and the whole stack, pipeline, database, API and portal, is built to come up on someone else’s machine just as easily as on mine. The code is licensed AGPL-3.0, a copyleft licence that keeps it open: anyone who runs a modified version as a service has to share their changes too. If MapleSpike vanished tomorrow, the method would not vanish with it. That is the difference between a tool and a tenancy, and State as Operating System explains why I am allergic to tenancies.

Building Beats Complaining

Build Canada² lists “Building beats complaining.” among its principles, and follows it with this: “Anyone can point at what’s broken; we’re focused on the actions needed for Canada to reach its true potential.” Their projects³ do exactly that on the accountability side, tracking hundreds of federal commitments from promise to completion and laying out what governments actually spend. Quill sits one layer lower. A commitment tracker tells you whether a promise moved. Quill hands you the committee transcript, the gazette notice and the contract award underneath it, with a fingerprint on each, so anybody, ’s tools or yours, can check the work instead of trusting the summary.

The public record already belongs to the public. It just has terrible indexing. Quill is an index with receipts, and the next question is yours.

Further reading: Quill¹, live · Build Canada’s principles² · Build Canada’s projects³, the federal commitment tracker and government financials.

Credit: the two lines on building and complaining are quoted verbatim from Build Canada’s About page². The line on hashes is quoted from the Quill site.

Quill ↩ ↩²
Build Canada — About. Their about page, in their words. ↩ ↩² ↩³
Build Canada — Projects. Their project list, including the public spending dashboards. ↩ ↩²