Agent State is the information that represents the agent's current execution state.
It can include things like:
- Current task or goal
- Conversation history
- Tool calls and results
- Current step or status
- Intermediate results
- Variables needed for the next action
For example:
{
"task": "Find the customer's latest order",
"current_step": "search_orders",
"customer_id": "123",
"tool_result": null
}The state changes as the agent executes:
State → Reason → Action → Observation → Updated State
State is usually temporary and execution-specific. It represents where the agent is right now, while memory is used to retain information beyond the current execution.
State answers: "What is happening in this execution right now?"