Cycles Guides
Long-form guides covering the two core enforcement dimensions of runtime authority. Each guide is a map — a short orientation per subtopic that links into the deep coverage. Read top to bottom for a structured view, or jump to whichever section matches what you are working on.
The two core enforcement dimensions of runtime authority
Cost and action are different enforcement problems. Cost controls how much an agent can spend; action authority controls what it is allowed to do. Most real production incidents touch both — a runaway agent that loops on the LLM is a cost incident; an agent that loops while writing to a database is an action incident; a single agent that does both is the canonical disaster scenario.
Tenant isolation and audit evidence build on these two enforcement dimensions: who owns the budget, and what record each decision leaves behind.
- LLM Cost Runtime Control Reference — bounding what AI agents spend. Causes of cost blowups, why dashboards are not enough, the runtime patterns that work, multi-tenant cost isolation, unit economics, and provider-specific patterns.
- AI Agent Risk & Blast Radius Reference — bounding what AI agents do. Risk scoring, action authority, blast-radius containment, degradation paths, delegation chains, governance frameworks, and incident patterns.
More guides on multi-tenant operations and audit / evidence are in development.