delta-agents
An AI agent framework built around a deterministic governance engine that validates, authorizes, and audits every agent action, using control theory and statistical estimation to keep agents from drifting from policy.
Delta Agents is a framework and runtime for building production AI agents with built-in safety and governance. Every agent operation runs through a deterministic, math-backed governance engine: the model proposes an action, and the engine validates, authorizes, supervises, and audits it, deciding whether it is safe to proceed, guiding the agent toward a correct action, or blocking it outright, with human-in-the-loop review available when the agent drifts from expected behavior.
The governance engine is built on control theory, decision theory, and statistical estimation, including bounded state-space models, Markov constraints, Bellman optimization, model predictive control, Kalman estimation, and Bayesian updating, and every decision is described as deterministic, provable, and auditable. Features include token, time, and multi-dimensional budget enforcement, schema and prerequisite validation before every action, risk scoring that gates high-stakes operations behind human approval, and loop detection that catches reasoning spirals before they burn budget.
Multi-agent teams can delegate work to each other under scoped budgets, communicating through mailboxes with delivery receipts and automatic recovery from failures.