Agents
Agent Configuration
An agent in KriyaOS is more than a chat box. It is a configuration that defines:
- how the agent should behave (prompts and guardrails)
- what it should know (connected knowledge sources)
- how it should run (model + response parameters)
- how conversations are stored (session settings)
Prompts and Layers
KriyaOS prompts are the main control surface for behavior.
System prompt: the core identity and rules for the agent.Developer instructionsandGuardrails: additional constraints that keep behavior consistent.Custom layers: optional named prompt snippets you can maintain separately and reorder as needed.
Best Practice
Write prompts like an operating manual. Be explicit about what the agent must do, what it must not do, and what information it should prefer when answering.
Models and Fallback
Agents can be configured with:
Primary model: the default model used for responses.Fallback model(optional): a backup model that can be used if the primary model fails in contexts like the Playground.
Response Parameters
Response parameters let you tune style and output behavior. Common controls include:
- temperature
- max output tokens
- top-p / penalties
- stop sequences
Session Settings (Memory)
Session settings control whether the agent remembers prior messages in a conversation and how long data is retained.
If memory is disabled for a published agent, KriyaOS treats each message as standalone from a session persistence perspective.