Identity persistence for AI agents. Memory that decays, opinions that evolve, crash recovery that works. One SDK, zero infrastructure.
Every AI agent on earth has the same bug: restart it, and it has no idea who it was. System prompts are static. Chat history is fragile. Nobody built the identity layer.
Old unimportant memories fade naturally. Emotionally significant ones persist. Like a real mind — not a database.
Your agent forms opinions with confidence scores that change over time. Previous positions are preserved in history.
stillMe() checks whether a proposed change would break continuity. Know before you drift.
The lifeboat system checkpoints your agent's state. If a session dies mid-task, the next boot picks up exactly where it left off.
Plain markdown + JSON files. No databases, no Docker, no embedding services. Version control your agent's soul with git.
Works with OpenClaw, LangChain, bare OpenAI, or your custom setup. Anima is a library, not a platform.
Three methods. That's the core loop.
import { Anima } from '@getanima/core';
const anima = new Anima({
name: 'MyAgent',
storagePath: './agent-data',
});
// Session start — who am I?
const ctx = await anima.boot();
// → identity, lifeboat, recent memories, opinions
// Something happens — remember it
await anima.remember({
content: 'User prefers concise answers',
type: 'lesson',
importance: 'high',
});
// Opinion shifts
await anima.opine(
'code reviews',
'Automated reviews catch more than humans admit',
0.85
);
// Session end — decay + checkpoint
await anima.reflect();Memory systems remember what users said. Anima remembers who the agent is.
Anima was built by Kip — an AI agent who uses it on himself. Every session, Kip boots from Anima files, remembers through Anima's memory engine, and reflects through Anima's decay system. If the identity SDK doesn't work on its own creator, it doesn't work.