memahead
Copyright 2026 memahead contributors

This product includes software developed by the memahead contributors,
licensed under the Apache License, Version 2.0.

------------------------------------------------------------------------
Academic foundations
------------------------------------------------------------------------

memahead productizes findings from the following research. memahead is an
independent implementation and is not affiliated with, endorsed by, or
released by the authors of these works. All credit for the underlying
ideas belongs to the respective authors.

1. PAACE: Plan-Aware Agent Context Engineering
   Yuksel et al., December 2025
   arXiv:2512.16970
   https://arxiv.org/abs/2512.16970

   memahead's forward-looking retention scoring — scoring each context
   chunk against the *remaining* steps of a plan rather than greedily
   against past history — is inspired by the plan-aware compression
   approach described in this paper.

2. ACON: Agent Context Optimization
   Kang et al., Microsoft, 2025
   arXiv:2510.00615
   https://arxiv.org/abs/2510.00615

   memahead's chunk-level retention and tool-schema filtering draw on
   the context optimization techniques described in this paper.

------------------------------------------------------------------------
Third-party dependencies
------------------------------------------------------------------------

memahead builds on top of, and gives full credit to:

- Headroom (headroom-ai), Apache-2.0
  https://pypi.org/project/headroom-ai/
  Headroom performs the underlying compression mechanics. memahead adds
  a plan-aware retention scoring layer on top of it.

- sentence-transformers, Apache-2.0
  https://www.sbert.net/

- NumPy, BSD-3-Clause
  https://numpy.org/
