About us

We build the governance layer that automated decisions need.

Codebeatz Technologies Inc. builds TovrisAI so a department can show, at any moment, which AI systems it runs, which rules apply to each one, and how it knows they are met. That answer should come from a record, not a scramble.

§01Our mission

Help every organization that uses AI to make decisions about people keep those systems aligned with the rules, with the reasoning visible, the evidence one click away, and a person accountable for every conclusion.

§02Our story

From a directive to a working loop

  1. The rules

    Clear obligations, manual work

    The Directive on Automated Decision-Making and its Algorithmic Impact Assessment set obligations by impact level. Applying them across dozens of systems is still mostly done in spreadsheets.

  2. June 2025

    The amendments

    The directive was amended: 11 requirements added, 34 modified, 6 retired. Each department had to work out what that meant for every system it runs.

  3. November 2026

    The build

    We built Tovris around one loop: register, encode the rules, decide what applies, assess, review, detect change. Each part is tested on a simulated Government of Canada environment built from the public AI Register.

  4. Today

    Validation

    We are measuring Tovris against expert labels for agreement, calibration, change detection and security, with an evaluation harness that ships in the product. Next: pilots with departments.

§03Principles

Four commitments we design against

Each one is enforced in the product, and each has a test that fails when it is broken.

  • P.01

    People stay accountable

    Tovris proposes; a reviewer decides. The model's answer is kept next to the reviewer's, so nobody has to guess who concluded what.

  • P.02

    Say what you don't know

    When evidence is missing or the samples disagree, the answer is “unable to determine”. A confident wrong answer is worse than an honest gap.

  • P.03

    Quote, don't paraphrase

    Requirements carry the exact wording of the official text in both languages, linked to the stored source. If a quote can't be found word for word, the build stops.

  • P.04

    Keep it in Canada

    Government of Canada data, and the model calls that use it, stay in Canadian regions. The AI tools we use to write code never see it.

Features in the Tovris specification
176
Modules, from registry to operations
22
Jobs in the name: Trace to Secure
6
Official languages, from day one
2
§04Work with us

Bring us a system you are responsible for

Departments that pilot Tovris work directly with the team building it, and their constraints shape what ships next.

In English or French · Synthetic data for every demo