Why Aegis

Aegis exists because most agents still fail in the same three ways:

• they forget what matters after the session ends
• they optimize for the current turn instead of the next week
• they overflow the context window and fall back to naive truncation

The bet behind Aegis is that a useful agent should behave more like a durable collaborator: it should remember, resume, and recover the right context instead of starting over every time.

What Aegis is trying to build

The public Aegis path is intentionally narrow. The supported product path today is:

1. install locally
2. create or confirm identity with aegis init
3. verify readiness with aegis status
4. create a named clone
5. continue the same clone through aegis wake

The public stack

• Installer: prepares the local runtime and launcher
• Launcher: exposes the aegis command
• Personal Model: stores long-term personalized context and identity
• State: keeps continuity data across wake windows and conversations
• CLI runtime: keeps the first-use, provider, clone, and wake flows in one shell

Why the public docs stay small

• The docs in Philosophy, Getting Started, Capacities, Reference, and Help describe the supported operator path.
• The public site should stay focused on what operators can actually install, run, and troubleshoot today.

Where system design fits

The repository keeps canonical system design under docs/system-design/. That design is intentionally narrowed to the current layer model instead of serving as an archive of historical drafts.

Use it when you want to understand:

• the Step -> Loop -> Episode -> State -> Personal Model hierarchy
• how wake windows resume durable State
• how explicit memory capture and grounded reflection are governed
• how Skill Affinity lives inside Personal Model Procedural Memory
• how skill crystallization promotes repeated successful workflows as a downstream expression of the Personal Model

Start with the repository system-design docs when working from a checkout.