AI Agent Configuration
A set of agent behaviors that can be toggled during a session. Each behavior specifies goals, tone, constraints, and switching commands. Behaviors can be combined, and the most recently activated behavior takes precedence on conflicts.
→ Full Specification#
→ Reference Index#
A comprehensive configuration covers:
| Feature | Description |
|---|---|
| Agent | Agent-specific settings, such as name, description, persona, and appearance. |
| Prompts | System, developer, and task prompts that guide the agent’s tone, scope, and response style. |
| LLM connectors | Mapped providers, models, parameters (e.g., temperature, max tokens), and instruction sources. |
| RAG integrations | Connectors to vector stores, retrievers, indexing policies, and grounding strategies. |
| Behavior and policies | Rules for safety, confidentiality, escalation, tool usage, and fallback logic. |
| Appearance and UX | Name, description, persona, response formatting, and icons and color themes |
| Secrets and credentials | Securely referenced tokens, API keys, and connection strings |
| Metadata | Author, version, changelog, licensing, and compliance notes. |
| Testing and validation | Unit tests, prompt regression tests, evaluation metrics, and monitoring hooks. |
| Operational settings | Rate limits, caching, logging, observability, and error-handling policies. |
| Last updated: | 2025-09-28 |