Architecting the
autonomy stack.
Three production codebases. One operator. VIGIL is the active bet, the SIEM for AI agents. GRiiD and CRYpT are open-source studio work, now archived to community maintenance.
Built solo. Built honest.
Built for the operator on call.
COS TRINITY is an Indigenous-owned software studio in Regina, Saskatchewan. We build operator-grade software for the AI agent era. Observability, compliance mapping, and the audit layer that lets teams deploy autonomous systems without inheriting their risk.
The studio shipped three production codebases in three months. The first two taught us what high-frequency execution audit logs and local-only privacy architectures actually look like in practice. The third, VIGIL, is the twelve-month bet that those lessons land in a category-defining product.
Three codebases. One focus.
VIGIL
The SIEM for AI agents
Real-time event feed, threat detection at ingest, compliance-native exports for SOC 2 / PCI / ISO / NIST / GDPR / India DPDP. Built for the operator on call.
GRiiD
Solana trading primitives
Self-hosted execution toolkit: grid, DCA, whale-copy, sniping. Shipped on mainnet under MIT. Community-maintained.
CRYpT
Local-only file encryption
14-cipher AEAD cascade with double Argon2id key derivation, 192-round Feistel network. Open source, no cloud, no key escrow.
Three months. Three codebases. One arc.
CRYpT
Local-only file encryption. Shipped what privacy primitives feel like when there's no cloud and no escrow. Set the "your data never leaves your machine" rule that every later product inherits.
GRiiD
Solana trading primitives on mainnet. Taught us what operator-grade telemetry needs when an automated process is moving real money in real time. Audit-log discipline carried directly into VIGIL.
VIGIL
The SIEM for AI agents. Where the prior two codebases' lessons compose: local-first architecture from CRYpT, audit-log discipline from GRiiD, written for the operator on call when an agent runs hot at 3am.
Operator-first.
Build honest.
Ship narrow.
Three things the studio holds to. The product gets built for the human on call, not the team that deployed it. The marketing says what shipped, not what's planned. The roadmap stays narrow because focus is the scarcest input at this stage.
Operator-first
The person on call when an agent burns through API credits at 3am sees a different problem than the ML engineer who deployed it. Tools built only for the engineer leave the operator holding the pager.
Build honest
Every claim on this site is what's true today. VIGIL's Aadhaar detector ships with a measured ~10% false-positive rate from Verhoeff alone, and we explain why that's a mathematical ceiling, not a bug. The fix (context gates) is a Phase 1 line item, not a marketing promise.
Ship narrow
Three production codebases proved we can build broadly. VIGIL gets the next twelve months alone. GRiiD and CRYpT are archived to community-maintained status. Velocity is a moat only when paired with focus.