Framework and Prior-Art Research

This section examines related agent frameworks and methodologies one by one and side by side. It contains the prior-art overview, framework comparison matrix, and detailed research pages for each framework.

Comparative Research

• Prior-Art and Comparative Research Overview — Summarizes the main areas of overlap, the strongest current ideas about what may be different, and other areas of prior art that still need investigation.
• Framework Comparison Matrix — Provides a principle-by-principle similarity comparison across the frameworks currently under review.

Framework Research Dossiers

• GenericAgent — Examines autonomous execution, layered memory, skill creation, reusable SOPs, executable capability creation, and compounding reuse.
• Metis — Examines the choice between textual and executable memory, including moving recurring procedural plans into callable code.
• MOSS — Examines production-failure-driven source changes, candidate replay, human-gated promotion, health checks, and rollback.
• MASFly — Examines SOP-driven multi-agent system construction, Watcher-based execution supervision, runtime intervention, and accumulated collaboration experience.
• Ralph — Examines repeated autonomous coding-agent loops, persistent repository assets, deterministic backpressure, and human loop engineering.
• Reflexion — Examines feedback-driven self-reflection and persistent reasoning memory used to improve future attempts without changing model weights.

Research Standard

Each framework research page should include, where available:

• the main paper, official site, or primary defining source,
• an official implementation repository,
• the earliest verified public publication or release date and the basis for that date,
• additional recommended reading relevant to comparison,
• a detailed description of the framework and its architecture,
• its execution, learning, persistence, validation, and governance mechanisms,
• a completed evidence checklist,
• direct comparison with the Infoconex AI Flywheel,
• and remaining questions requiring more primary-source research.

The research prioritizes original papers, author-maintained project pages, official source repositories, and documentation published by the framework authors.

For the principle-oriented research view, see the principle research dossiers.