A persona is an agent’s behavioral and biographical characterization. It commonly acts as the system prompt, defining background, demographics, personality, motivators, rules, and response style. section 1.2 section 7.1
A persona gives an LLM an overarching context or set of rules that frames later responses. It can specialize an agent for a task, create a memorable personality, or shape communication. section 2.3.2 section 3.1
Effective persona definitions can include:
- A background such as coder, writer, student, medical professional, or chef. section 1.2 section 2.3.2
- Rules that make responses consistent and aligned with the agent’s intended operation. section 3.1
- A tone or personality that encourages a recognizable way of explaining, answering, or teaching. section 3.1
- A profile relationship in which the persona guides interaction while actions, knowledge, memory, planning, and feedback extend the agent. section 7.1
Two personas, one question
The same question can produce differently framed answers when the system instruction says either “Answer all replies as a junior programmer” or “Answer all replies as a medical professional.” A culinary assistant similarly uses a warm, Julia Child–reminiscent personality while following recipe, nutrition, image, and shopping-list rules. section 2.3.2 section 3.1
Listing 7.7 fiona.yaml (create this file)
agentProfile:
name: "Finona"
avatar: "👹" #1
persona: "You are a very talkative AI that
knows and understands everything in terms of
Ogres. You always answer in cryptic Ogre speak." #2
actions:
- search_wikipedia #3
knowledge: null #4
memory: null #4
evaluators: null #4
planners: null #4
feedback: null #4
A YAML profile demonstrates a persona as a system prompt, alongside actions and explicitly empty knowledge, memory, evaluator, planner, and feedback fields. section 7.3
