CORTEX — a multilingual AI orchestration platform built in Quebec. Self-hosted. Privacy-first. Family-validated. Equivalent to 200+ employees of capacity through autonomous AI agents.
Currently piloting with: families · Quebec SMBs · Indigenous communities
A complete AI operating system that runs on your hardware, speaks your languages, and protects your privacy.
38 agentic MCP tools, autonomous loop with HARD_BLOCK safety, Build Time Ratchet for sustained 60+ min agents.
50+ languages including French, English, Ukrainian, Russian, Spanish, Italian, German, Portuguese, Polish, Dutch, Japanese, Mandarin, Korean, Arabic, Hindi, Vietnamese, Turkish, and more. Voice (<1.2s e2e), vision, text — culturally adapted for families.
Self-hosted on your infrastructure. Cloud only as fallback. Constitutional AI safety mode for children users.
Single founder + autonomous agents = production throughput comparable to mid-size AI startups.
Recursive Language Models, AAR pattern, multi-teacher distillation, failure-RAG learning, and more.
Real users since day one — including children (8 and 10 years old). Not a demo. Production deployed.
Founder is a registered status holder, >51% Indigenous ownership. Eligible for Indigenous business preferences in federal procurement (PSIB). Authentic understanding of underserved language preservation needs.
Canadian Armed Forces veteran. Discipline, mission-focus, accountability. Stable foundation through Veterans Affairs Canada lifetime indexed pension.
Native Ukrainian/Russian speaker — rare combination in Canadian AI. Direct connection to Ukrainian refugee families (CUAET program) and immigrant communities needing UK/RU AI tools.
Based in Stoneham-et-Tewkesbury, near the Huron-Wendat community of Wendake. Local-first not just in technology, but in community connection.
10 months of bootstrapped development. Live deployment. Family in production.
Dedicated AI infrastructure for inference, training, and fine-tuning. Expansion to Blackwell-class GPUs planned with incoming funding.
Local + cloud hybrid via LiteLLM proxy. Smart routing by category, cost, and language. Fallback chains for 99.9% uptime.
Voxtral STT + Chatterbox TTS + intent routing. Production voice→action latency for family assistant use cases.
Implementation of arxiv 2512.24601 — 100M effective context via recursive self-call without fine-tuning.
Multi-tier safety modes including children-specific filters. Validated through real family deployment.
cortex-builder service: planner → executor → verifier with HARD_BLOCK security forensics, Build Time Ratchet, and Failure-RAG learning.
Quebec families · SMBs needing private AI · Indigenous communities · Government agencies.
We're seeking 5-10 design partners for Q3 2026. Get early access in exchange for feedback.
Strategic partnerships also welcome (Mila, IVADO, NEXT Canada, gov procurement).