Digby vs Augure AI

Augure AI — operating as The Altercation Company in Toronto — offers models it has trained itself and named for Canadian places: Rosedale One for frontier reasoning, Ossington 5 as a Canadian-tuned model for law, tax and Quebec French, and Tofino 3 for fast everyday work. Their pages state storage in Canada and inference in Canada under zero-retention agreements, never through US providers.

Rosedale One now has a page of its own. They describe it as a frontier reasoning model built by post-training an openly released frontier foundation model — Quebec French, Canadian legal and civic context, and release gates — and they serve it themselves, so prompts go to their Canadian gateway rather than to a third-party model API. They also started publishing numbers a buyer can check: 98.2 on HumanEval (n = 164, 95% CI 94.8–99.4), 86.7 on AIME 2025 (n = 30, 95% CI 70.3–94.7), and 100.0 on grounded answers from Québec sources (n = 60, 95% CI 94.0–100.0), plus $6 CAD per 1M output tokens on their developer API. Publishing a sample size and a confidence interval alongside a score is more than most vendors do, and it is worth crediting.

They make their ownership explicit in a way that matters to Canadian buyers: no US corporate parent and no US investors. Pricing is published in CAD on an annual view — free, Pro at $200 a year and Max at $800 a year, with a monthly view and a badge for two months free — and their French FAQ still quotes the monthly ladder plus an Ultra plan for all-day coding use. Enterprise options cover custom VPC, on-premises and GPU deployments, and they run a public trust centre on Vanta.

That is a real sovereign Canadian product, not a rebadged foreign API. They also ship more than chat: a knowledge product, contract review for Canadian law, and a coding agent of their own. If your requirement is “a Canadian model, from a Canadian company, with Canadian inference”, they satisfy it honestly — and on the specific question of published model benchmarks, they are ahead of us today.

Sources Augure AI — models, sovereignty claims and team · Augure AI — Rosedale One model page, benchmarks and API price · Augure AI — French pages and FAQ (monthly plans)

1. 01

   Open weights

   Digby runs custom-tuned open models rather than proprietary ones, so the intelligence is inspectable and replaceable, and a better open model can be adopted when it ships.

2. 02

   Multi-model routing plus a check

   the answer comes from the model suited to the question, and contested or high-stakes answers get an independent second model reviewing the first.

3. 03

   Deep research

   a council of models researches a question in parallel and writes one sourced report.

4. 04

   A coding agent in the same product

   Digby Code works in a real workspace and opens pull requests.

5. 05

   No-account start

   open digby.ai and type, rather than choosing a plan first.

6. 06

   Self-host on anything

   the full stack runs against any OpenAI-compatible endpoint: vLLM, Ollama or LM Studio on your own GPUs.

Choose them if your procurement rule is that every layer — model, company, storage, inference — must be Canadian; if you want a published per-seat price list and published model benchmarks to compare against a budget or a technical bar; or if you want a vendor whose headline promise is a named model rather than a routing layer.

Choose Digby if you want to start in ten seconds without a plan, if open weights matter to your exit strategy, if you want the research council and coding agent in the same product, or if your path ends in self-hosting on hardware you already own.

1. 01

   Where is the data stored, and where is the prompt processed — can you name both?

2. 02

   Are the models open-weight, and can we swap them if a better one appears?

3. 03

   What happens to our data if we leave?

4. 04

   Can we run the whole system ourselves if we need to?

5. 05

   Do you hold any certification or audit report we can read?