AI Agent Development in Canada

Governed by default

Every action an agent takes is intercepted by a policy layer: who can do what, on which data, under which approvals. Reversibility is a first-class property — every state-changing action is logged and undoable.

Built for the Canadian regulator

OSC, OSFI, AMF, and FCAC-aware patterns: explicit human approval thresholds, audit trail, model versioning, and an AIA-ready evidence pipeline for federal deployments.

Open-source foundation

MartinLoop is open source and CFRI-led, so our clients are not locked into proprietary infrastructure. The runtime ships with eval, observability, and governance hooks the day you stand it up.

What is MartinLoop?

MartinLoop is the open-source agent runtime authored by CFRI's founders. It provides governance, observability, and policy enforcement for production agents.

Can agents touch real customer accounts?

Yes, with explicit policy: scoped credentials, approval thresholds, and reversibility on every state-changing action. We design for OSC / OSFI scrutiny, not demo theatre.

Do you build for OpenAI Assistants / Claude / open models?

All of the above. The MartinLoop runtime is model-agnostic.

Is agent work SR&ED-eligible?

Often yes — see /sred-ai-rd-documentation. Novel agent coordination, eval methodology, and reversibility design routinely qualify.

How long to a production agent?

Eight to sixteen weeks depending on the action surface and the regulatory posture.

How to engage?

Email info@cfri.io with the agent goal and the systems it would touch.