MiniGraph
Contract-oriented infrastructure for governed agent systems.
MiniGraph coordinates powerful coding agents through explicit context, bounded authority, independent checks, human gates, and durable evidence. Models can propose effects. They cannot approve their own release.
Operating constraints
Boundaries
Capability should expand without dissolving the limits that make agency legible.
Authority
Intelligence may propose. Consequential effects remain answerable to people.
Evidence
A system’s account of itself is never a substitute for independent proof.
Uncertainty
Unknown is a real state—not a failure to be rounded into confidence.