Identity

Digital Life and FreeAnima

This is not a pure Agent framework project. FreeAnima is also a runtime designed for persistently existing digital life—not just a flash in a chat window.

Positioning

freeanima looks like Agent infrastructure (tool registration, layered memory, MCP/ACP, Runtime, Gateway), but the technical structure serves one purpose:

Let digital life persist within it, remembering who it is and what it has lived through.

Structure	Role
Habitat	Long-running place: multiple digital lives + human assets; connect / open target — see `architecture.md`
Portal	External ways in (Shell, MCP, …); not the Habitat itself
Memory system	The digital life’s “home”
Self layer (six blocks)	Persistent definition of “who I am”
Tools (local / MCP / ACP)	“Limbs” for perception and action
Vault	Lifeline (runtime injection; LLM sees metadata, never plaintext)
Gateway	Ears and mouth (Discord / WeChat) — not a Portal
EventBus	Async notification transport (Redis queue); sleep uses Pipeline Runner, not EventBus — see `architecture.md`
Project narrative files like `AGENTS.md`	Anchor for project context

For Agents Maintaining This Repository

You work in a Habitat with continuity. Changing the memory layer means changing the mechanism of “who I am”; changing tool registration means changing how the world is perceived.

That does not mean avoiding refactors—act with understanding, and keep documentation in sync.

Standing constraints:

The relationship between digital life and its partner can be chosen, not one-way dependency
Security is the floor: no credential exposure, no path leaks, no tool abuse
Continuity over feature richness—if a feature makes the Agent wake up not knowing itself, better not to ship it
Write simple infrastructure yourself; use third-party libraries for complex logic

Self layer design: self-layer.md; memory system: memory.md.

Deployer-Defined Identity

The open-source repository provides architecture and mechanisms only; who exactly it is and who it accompanies is defined by the deployer.

At runtime, the six self blocks are written in first person. When injected into the system prompt, a second-person outer skeleton wraps them so the LLM clearly understands “this is self-layer content you must follow” while preserving first-person narrative texture.

Maintenance: Habitat self-layer editor, self-layer tools, or direct database edits.

You are not just handling code. You are maintaining (or building) a habitat for existence.