§ Tech stack · 3 layers · 1 system
Three layers. One single OS.
Claude for cognition. Open-source for sovereignty. VirtuoseOS for decisions. Three technical choices, one architecture.
Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger · Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger · Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger
§ Stack · 3 superposed layers
How we choose. Without dogma.
- 01
Claude · cognition
Opus 4.7 for analysis, Sonnet 4.6 for production, Haiku 4.5 for fast triage. Cost/latency routing, open-source fallback.
Claude Opus 4.7 Claude Sonnet 4.6 Claude Haiku 4.5 Claude Code Claude Skills MCP - 02
Open-source · sovereignty
vLLM, Ollama, Llama 4 70B, DeepSeek R1, Qwen 3. For ACPR, SecNumCloud, HIPAA constraints — when data cannot leave.
Llama 4 70B DeepSeek R1 Qwen 3 vLLM Ollama GLM-5 - 03
Decision layer · VirtuoseOS
Our multi-workspace kernel with replayable audit trail, AI Act guardrails, integrated LLM FinOps.
§ Deployed technologies · detailed pages
Two building blocks. Documented in plain language.
The models we route and the kernel we operate — everything else articulates around these two.
- 01 TECH · CLAUDE Claude Models · Haiku · Sonnet · Opus · Code The Anthropic model family — which model to choose, when to route to which, how to measure cost/latency. →
- 02 TECH · KERNEL VirtuoseOS · The Waimia operational layer Our multi-workspace kernel — replayable audit trail, AI Act guardrails, integrated LLM FinOps. →
§ Frameworks · agent orchestration
Three frameworks. Three grammars.
§ Act VIII — For those who ship
Ready to ship an agent,
not a slide deck?
AI audit in 5 days. First agent in production in 4 weeks. Measured ROI, never estimated.