Open-source research

PseudoAgent

A minimal, state-driven agent harness implementing the PseudoPlan-Decide-Act-Digest (P2DAD) loop. Agents are defined in YAML, with explicit decision criteria and state updated through digests.

Tool selection. The default implementation uses decision models together with an Analytic Hierarchy Process (AHP) decision support system to choose tools. This design is intended to increase speed and make the harness usable with less capable models that may struggle to select tools reliably through native tool calling.

AHP is an implementation choice, not a requirement of PseudoAgent. Other implementations can use the model's native tool-calling mechanism to select tools within the same state-driven loop.

https://github.com/pseudoconscious/pseudoagent

Independent research supported by PseudoConscious Inc.