Recension Research builds and evaluates AI systems around private and local inference, agent runtime instrumentation, provenance, permissioned execution and reproducible verification.
Map workflows and data boundaries, compare hosted versus private deployment, identify security and permission risks, and leave with an implementation plan rather than an AI strategy deck.
Instrument an existing agent, build reproducible failure tests, examine permissions and tool access, and turn “it worked in the demo” into evidence a team can actually inspect.
Prototype local or private model deployments, bounded agent workflows and evaluation harnesses around real operating constraints — with documentation and a clear handover path.
The public projects explore a recurring question: how do increasingly capable agent systems remain inspectable, evidential and governable without pretending one mechanism solves everything?
Instrumented cognition for coding agents. Observe and steer runtimes through explicit semantic contracts, compare experiments, and keep evaluation separate from promotion.
A replayable memory kernel for AI systems: signed history, explicit evidence, provenance, deterministic replay and policy-defined conclusions.
A permission engine for AI-operated systems. Models may propose actions; typed routes, policy, leases, confirmation, rollback and receipts keep authority outside the model.
A Rust recorder and verifier for agent runs, sealing committed run semantics under tamper-evident witness roots that can be independently recomputed.
Recension Research is an independent AI systems research and engineering practice founded by George Gallagher.
The commercial work starts with deployment problems — privacy, agent reliability, auditability, permissions and evaluation — while the open-source research develops reusable mechanisms for evidence, control and verification. Consulting is useful when it discovers the product; research is useful when it survives a real constraint.