System prompts
A system prompt defines an agent's role, desired outcome, conversation policy, trusted information, action rules, and boundaries. Wirevox compiles that authored policy with runtime context before each conversation.
Use this topic to design behavior that remains understandable, testable, and safe as the agent gains models, functions, variables, and knowledge.
In this topic
- Write an effective system prompt — use the Wirevox prompt contract, adapt it to the selected model family, write complete function rules, and test the result.
- Variables and context — keep stable configured facts separate from live conversation context, knowledge, and action results.
- Functions and integrations — connect capabilities, attach them to agents, write exact action rules, and test external outcomes.
- Knowledge bases and RAG — index approved source material, choose retrieval scope, and define grounded fallback behavior.
Future focused guides will cover reusable templates and prompt evaluation without turning one page into an oversized reference.
Recommended sequence
- Choose the agent's model and pipeline architecture.
- Write the smallest complete prompt for the intended outcome.
- Add variables, knowledge, and functions only where the workflow requires them.
- Test expected, ambiguous, failure, and escalation paths.
- Publish a named version and monitor real conversations.