Runtime-first agents
Agent behavior is routed through durable runtime capabilities instead of leaving every task as one-off chat output.
Architecture
Runtime-first agents with typed, inspectable execution.
Buffaly separates language reasoning from executable runtime capability so high-trust workflows can be reviewed, constrained, and connected to real systems.
Typed actions
Actions expose explicit inputs, outputs, and contracts so execution can be inspected and validated.
Semantic entities
Important nouns become typed objects the runtime can bind to, query, reuse, and refine.
Prototype graphs
Prototype-based structure lets runtime behavior, domain knowledge, tools, and prompts compose as inspectable graph objects.
Inspectable execution
Plans, scratch notes, typed tool calls, evidence, and validation steps make behavior reviewable instead of opaque.
Domain packs
Healthcare and other high-trust domain packs layer on top of the public core runtime and remain separate when they include sensitive workflows, connectors, or deployment playbooks.