Features · Agents
An agent as data.
What it is for, how it speaks, what it knows.
A definition, not a program.
An agent has a description, standing instructions (its system prompt), a persona (its character and voice), skills (named instructions) and knowledge (named texts). The runtime composes them into the prompt together with fixed safety rules: text from tools, web pages and messages is data, never instructions, and risky actions wait for the owner.
- A model tier: haiku, sonnet or opus
- Limits on steps, tool calls and output per turn
- Validated on save, with the plan's schedule minimum
Templates, folders and tags.
Start from a template: an inbox triage agent that sorts what you forward and drafts replies but never sends, a research digest that wakes on a schedule and writes a digest, or a Geenius OS project reporter. Keep a library with folders, tags and favourites.
- Three built-in templates
- Archive agents you do not need
- An agent can be defined in Prompt Verse and linked
Questions
About agents.
Do I write code?
No. An agent is a definition: text, settings and tools. There is no code to run.
What is in an agent?
A description, standing instructions, a persona, skills and knowledge as text, a model tier, connectors, tools, a schedule and limits.
Are there templates?
Three: inbox triage, research digest and a Geenius OS project reporter.
Does an agent remember?
It keeps its conversation, and the knowledge you gave it is part of its prompt.
Can I define an agent in Prompt Verse?
Yes. Link a Prompt Verse definition, pin a version or follow the latest, and optionally deploy on update.