Models
Piccolo & Bozza
Explore our open-weight models.
Independent Canadian AI
Advancing AI to solve difficult problems.
We develop AI models, engineer the technologies around them, and build products for real work. Discover Vinci by SimpleDirect®.
Intelligence. Built into something useful.
Explore our models, the technologies behind our work, and the products that put AI in people’s hands.
Models
From our first model to the next generation. Discover the ideas, foundations, and trade-offs behind every release.
Technologies
The engineering behind Vinci: coordinating compute, governing work, and examining what the evidence supports.
Products
Tools for doing the work. Start with Vinci Code CLI and explore the interfaces we are continuing to develop.
Selected work
Public project capture, preserved from the software registry. It illustrates this recorded workflow, not every possible outcome.
Explore the work behind Vinci.
Software
Vinci Code CLI
A coding agent for your terminal.
Research and safety
When passing isn’t enough
Research into what AI evaluations miss.
Accountability in software
Vinci Contracts
Define the work. Record the evidence.
A real run, with its limits in view
This existing public recording shows a coding turn. An outcome can remain unverified when repeatable checks are missing. Vinci Code builds on Pi; the project page documents upstream attribution and the scope of its controls.
Demonstration, source, and limitationsWho the work is for
Different organizations. Concrete questions.
These are the audiences we seek to work with. Any collaboration starts with a discussion of fit, scope, and availability.
Small businesses
How can a small team complete more useful work without creating an unmanageable technical burden?
Explore collaborationEnterprises
How can AI participate in engineering and operational workflows with reviewable outputs and bounded access?
Explore collaborationGovernment and public institutions
How can public institutions assess AI systems, inspect their limitations, and explore accountable uses?
Explore collaborationThe research record
Results that can be examined.
We study AI-model behaviour and the reliability of the tests used to judge it. Technical reports preserve methods, trade-offs, corrections, and negative findings.
- Runtime Pass Is Not Correctness
- Character Transfer Across Three Model Families
- Transferring Character Post-Training to Mistral 7B
Our mission
Building an independent Canadian AI institution.
SimpleDirect® develops AI models, technologies, and products. Our work brings research and engineering together in Toronto.
Our purpose and peopleExplore SimpleDirect®
Start with the resources.
For individuals, businesses, and governments: explore our public models, technologies, product updates, and documentation. Contact us if you have further questions or want to work together.
Our next products will be unveiled as they become available.