User or agent intent
A request enters Maple with identity, workspace, budget, and consequence class.
Maple AI OS
AI-native operating systems
Frontier models are powerful, but they are not operating systems. Maple AI OS treats models as proposal engines and puts durable control around them: identity, capability gates, deterministic kernels, WorldLine receipts, rollback, evaluation, and promotion policy.
6tutorial pages
Operating-system builders, agent platform engineers, enterprise AI operators, and researchers studying reliable autonomy.
Thesis
The winning AI-native operating system is not the largest model. It is the layer that turns model uncertainty into authorized, inspectable, reversible system change.
This domain hub is designed for two jobs at once: educate serious users and create better training direction for Hermon adapters. The same vocabulary appears in tutorials, demos, eval rubrics, and model output contracts.
Architecture Map
MapleAI authority loop
The OS owns identity, policy, execution, evidence, and promotion. Hermon OS drafts typed proposals inside that loop; it does not own authority.
Hermon OS proposal
The model returns structured intent, actions, rollback, receipts, and denied operations.
Maple policy kernel
Deterministic checks decide capabilities, approvals, model route, and fail-closed behavior.
Execution and WorldLine
Approved services run, receipts record evidence, and evals decide the next promotion.
Research Questions
The questions this domain must answer
What should be deterministic when the intelligence layer is probabilistic?
How should agent identity, memory, tools, and budgets survive model upgrades?
What evidence is sufficient for an operator to trust an autonomous action?
How can local, cloud, and domain models share the same authority boundary?
Core Concepts
Concepts users need before trusting the demo
Proposal is not authority
Hermon OS may draft an action, but Maple kernels decide whether the user, agent, policy, budget, and current state authorize it. This is the central difference between a chatbot and an AI-native OS.
Agents are supervised processes
An agent needs an identity, memory scope, lifecycle state, capability set, schedule, health signal, budget, log stream, and receipt trail. Those are OS concerns, not prompt engineering details.
Receipts create continuity
WorldLine receipts preserve intent, policy checks, approvals, execution result, rollback state, and model version so work can be audited across provider failover and model promotion.
Evaluation becomes admission control
A model adapter should not go live because it trained longer. It should go live because it passed the contract probes that represent the domain's real work.
Tutorial Series
Work through the curriculum
Each lesson includes foundations, applied architecture, a lab prompt, evaluation checks, and an output contract that can become training data or demo validation.
Foundation25-35 min
The OS as an authority layer
Why AI-native systems need a deterministic authority layer around probabilistic models.
Open lessonFoundation25-35 min
Agent lifecycle, memory, and state
How Maple treats agents as supervised processes with scoped memory and lifecycle controls.
Open lessonApplied40-60 min
Capabilities, tools, MCP, and A2A edges
How an AI OS should expose tools to agents without turning every connector into a vulnerability.
Open lessonApplied40-60 min
WorldLine receipts and proof-carrying work
How Maple uses receipts to make autonomous work attributable, replayable, and verifiable.
Open lessonAdvanced60-90 min
Model routing, evals, and promotion gates
How Maple decides when a Hermon adapter is ready to serve a real domain.
Open lessonAdvanced60-90 min
A Canadian AI infrastructure strategy
How MapleAI can compete by building trusted AI infrastructure rather than chasing frontier-model scale.
Open lessonIndustrial Map
Where this becomes a product or operating capability
Maple AI OS can govern agents that touch files, SaaS APIs, customer records, schedules, code repositories, and financial workflows.
Maple AI OSGPT, Claude, Gemini, Hermon, Ollama, llama.cpp, and future local engines can be swapped without losing policy, memory, identity, or receipts.
Maple AI OSAgents, policies, prompts, drivers, eval packs, and model routes become signed artifacts with versioned deployment and rollback.
Maple AI OSHealth care, finance, legal, and public-sector AI need explicit commitments, review queues, no-action states, and proof of what happened.
Maple AI OSReading Map
Primary references and operational context
BIS FSI: Regulating AI in the financial sector
Regulatory context for AI risk, operational resilience, and financial-sector supervision.
Financial Stability Board: The Financial Stability Implications of Artificial IntelligencePolicy context for AI governance, vendor concentration, model risk, and financial stability.
MapleAI demosCurrent public eval snapshot for OS, DNA, Finance, and Code adapters.
Turn this domain into model improvement
Use the tutorials as user education, data-generation prompts, eval design, and demo scenarios for the next Hermon adapter cycle.