Agent Loop
The agent loop coordinates model calls, tool calls, tool result feedback, and final answer generation.
Core Objects
| Object | Purpose |
|---|---|
ChatRequest |
Agent input |
ToolCall |
Model-requested tool call |
ToolResult |
Tool execution result |
LLMResult |
Provider response |
AgentResponse |
Final agent response |
Finance-specific validation and harness logic should stay outside the generic loop and be implemented through tools, harnesses, presenters, or plugins.