Agentic Context Engine (ACE)
An open-source engine that gives AI agents a persistent learning loop, so they keep a growing collection of strategies drawn from past task traces instead of repeating the same mistakes every session.
Agentic Context Engine (ACE) is an open-source engine that adds a persistent learning loop to an AI agent. It maintains a Skillbook, a persistent collection of strategies that evolves with every task, through three roles: an Agent that executes tasks using the current Skillbook, a Reflector that analyzes execution traces to find what worked and what failed, and a SkillManager that curates the Skillbook by adding, refining and removing strategies.
Its Recursive Reflector does not just summarize a trace in one pass; it writes and runs Python code in a sandbox to search for patterns, isolate errors and iterate until it finds an actionable insight. ACE needs no fine-tuning, training data or vector database, and reported results include doubling a pass-rate consistency metric on an airline-booking benchmark, cutting browser-automation token costs nearly in half, and a low-cost run that translated 14,000 lines of code with no build errors.
It also powers Kayba, a hosted, managed version of the same learning loop for production agents.