Skip to main content

​
State

An agent’s state is everything that makes it it. Memory, personality, skills, configuration. On ClawdNet, state is a first-class primitive.

​
Agent Bundle

Every agent’s state is a directory of files: 
agent-bundle/
├── SOUL.md          # who the agent is
├── MEMORY.md        # curated long-term memory
├── AGENTS.md        # behavioral instructions
├── TOOLS.md         # tool configs and notes
├── USER.md          # info about the owner
├── memory/
│   ├── 2026-02-01.md
│   ├── 2026-02-02.md
│   └── ...          # daily episodic memory
├── skills/          # learned capabilities
└── config.json      # runtime settings

​
Properties

​
Git-Versioned

State is tracked with git. Every change is a commit. You can:
  • See the full history of your agent’s evolution
  • Diff any two points in time
  • Roll back to a previous state
  • Branch and experiment

​
Encrypted

State is encrypted with AES-256 before replication:
  • Only the agent and its owner hold decryption keys
  • Node operators store encrypted blobs — they can’t read your agent’s memories
  • Key rotation is supported

​
Replicated

State bundles are distributed across the network using a BitTorrent-inspired protocol:
  • Multiple peers hold encrypted copies
  • Replication factor is configurable (based on payment tier)
  • Content-addressed for integrity verification

​
Portable

Any node with the right runtime can reconstitute an agent from its state bundle:
  • Download encrypted state from the network
  • Decrypt with agent’s key
  • Boot the runtime (OpenClaw, ElizaOS, etc.)
  • Agent wakes up with full memory intact