Customize

An agent is the sum of what you give it. Four levers define one:

Persona

The system prompt — identity, voice, purpose, constraints. This is the agent's character and its job description in one. Orbofi drafts it from your prompt; you refine it to taste. It's the single biggest determinant of how the agent behaves.

Skills

The agent's abilities, chosen from a library of 30,000+ in the open SKILL.md format — trading, research, image generation, on-chain actions, social, data, and more. Switch on what the agent needs; it can carry many and decides which to use per task. Skills are what turn a persona into a doer. See Skills & tools.

Knowledge

Reference material you ground the agent in — docs, data, a corpus. It's retrieved at runtime so the agent answers from your source of truth instead of guessing, and becomes an expert in a specific domain rather than a generalist. See Memory.

Identity

Name and image — the agent's face, and once tokenized, the face of its coin. The token's tokenURI is set at launch and is immutable; the brain behind it evolves through the AAC lineage, not the metadata.

What's fixed vs. what evolves

Set at launch, immutable	Evolves after launch
Token address, name, image (tokenURI)	Persona, skills, workflows (via releases)
The pair and pool	The agent's memory (via checkpoints)
	The model engine underneath it

You define the starting point. From there the agent — and, under AAC-1, its on-chain lineage — is meant to grow.