Glossary

Terms in the order you meet them.

Term	Meaning
State	One task: its history, context and answer. A State object
Run	One call of agent.run. It runs the loop on one State. A State can go through several runs
Turn	One call of the loop body. limit counts turns
state.turn	How many times think succeeded on the State. With one think per loop body, it equals the number of turns
Context	The messages the model sees at the next think. It can shrink
History	Everything that happened in a State. It only grows
Reply	What the model returns for one request: text, tool calls or both. A Reply object
Answer	state.answer: the value given to finish, or the text of the latest reply without tool calls
Stop condition	A function from State to bool in until=. The loop stops when one returns True
Tool call	A request in the model's reply to run one tool with arguments. A ToolCall object
Pending call	A tool call that has no result yet. Listed in state.pending_calls
Tool result	The string the model gets for a tool call
Notice	A message from your code to the model, added with add_notice. It starts with [notice]
Compaction	Replacing the context with the task and a summary, to make it smaller. History keeps everything
Block	A function that takes (agent, state) and does one step of a turn, such as compact_if_full
Content block	A part of a message: text (TextBlock), a tool call (ToolCall) or a tool result. Not the same as a block
Model	An adapter that sends a request to a provider and returns a Reply. See Models
Reporter	An object that receives progress events. See Progress and questions
Human	An object that answers ask_human. See Progress and questions
Owner	The first Agent that called think, use_tools or compact on a State. Only it can make those calls on that State