Architecture
Three clusters, one controlled here. The target architecture of the Operations cluster and the constitution that governs how the three cooperate without coupling.
The three clusters
Each agent owns a cluster and competes for resources on evidence. grimaldi.ca is the control surface of exactly one of them.
Agent 1 · Energy Intelligence
“Create the highest-value deep-tech company.”
Software that observes, predicts, simulates, optimizes, coordinates and eventually safely controls distributed energy infrastructure. GridOS, DERIM, Energie Teilen, NeuralBridge, physics-informed.
Not controlled from this site. Counterparty in the integration contracts.
Agent 2 · Physical AI & Robotics
“Prove whether physical autonomy can create a defensible second moat.”
One coherent physical-AI stack — perceive, model, plan, act, verify, recover, learn — proven on one economically superior workflow first. Palletizer, robot-lidar-fusion, autonomous inspection.
Not controlled from this site. Counterparty in the integration contracts.
Agent 3 · Operations & Commercial Automation
“Make money and build distribution.”
Turn existing operational software into a profitable, repeatable business. Primary KPI: revenue. Not a research laboratory. No generic SaaS infrastructure without a customer-driven reason.
Controlled from grimaldi.ca/ — this site.
Target architecture of the Operations cluster
Runway Fuel is the Operations OS — a hypothesis until two independent customers require the same workflow. Applications sit on top. The registry entries are placed on the layer they serve.
The registry, entry by entry → · The contracts to the other clusters →
Group constitution
The rules all three agents obey. Verbatim in substance.
- Clusters
- There are exactly three strategic clusters. No agent may create a fourth.
- Ownership
- Each cluster owns its domain. Domain logic remains isolated: energy logic remains Energy, robotics logic remains Robotics, operations logic remains Operations.
- Moves
- No repository moves between clusters without documenting reason, commercial benefit, technical benefit, migration cost, dependency impact, and the CEO approval requirement.
- Sharing
- Shared primitives only when genuinely cross-cluster: authentication, billing, observability, deployment tooling, AI model gateway, secrets management, analytics, audit logging, documentation tooling.
- Communication
- Clusters communicate through versioned APIs, events, schemas, contracts and documented interfaces — never through undocumented database coupling.
- Project gate
- CUSTOMER PAIN → BUYER → MONEY → DIFFERENTIATION → TECHNICAL FEASIBILITY → STRATEGIC FIT → EXECUTION COST. A proposed project that fails two or more criteria is archived.
- Artefacts
- Every agent maintains: repository registry, architecture map, dependency graph, decision log, kill list, roadmap, customer evidence, competitive intelligence, research backlog, weekly KPI report.
- Not rewarded
- number of commits, number of repositories, lines of code, number of features, architectural complexity.
- Rewarded
- revenue, customers, deployments, measurable ROI, technical benchmarks, proprietary IP, research quality, reduced engineering complexity.
- When uncertain
- Prefer smaller scope, fewer repositories, clearer ownership, faster customer validation, stronger interfaces, lower burn, higher evidence.
- Allocation
- Agents compete for resources on evidence, not activity. Whichever cluster produces contracted value gets the next euro and the next engineer.
Allowed repository statuses
Six. No other status exists; the typecheck refuses a seventh.