Project · Memory Utility Labs

AEGIS

A local-first agent operating system for AI coding work. Tokens are finite inventory: pack once, reuse while bytes are unchanged, land the result, ledger everything. If it is not in the ledger, it did not happen.

What it does

AEGIS meters the token supply chain of AI-assisted development. A kernel tracks process, memory, and tool calls; a portable data plane holds the ledger, content-addressed context packs, and outputs; a frozen /v1 API keeps the contract stable while the implementation moves. A multi-model router, a daemon, and a menu-bar app round out the working surface.

The honest-yield rule

Efficiency claims are the easiest thing in this field to fake, so AEGIS refuses to make them casually: a savings claim stays null until an admitted pair of runs proves it. No admitted pair, no percentage. The rule is the feature.

Protective reserve

The doctrine holds a protective reserve floor of at least 80% under a weekly token budget — austere by design, aggressive on drift. The reserve is what keeps a long session honest.

Status

v1.2.0, released 2026-09-27. Public on GitHub at github.com/marsojuji-cmyk/aegis. Published under the Memory Utility Labs imprint; the publication of record is Introducing AEGIS, Technical Specifications Vol. I (2026).

What is not claimed: adoption, users, yield percentages, or benchmark outcomes. The 30-day existence proof for the honest-yield pair is in progress; until it lands, there is no number here.