Cap historical tool calls to control context size and spend

A new parameter, max_tool_calls_from_history, lets you cap how many historical tool call pairs are loaded into context. This keeps token usage predictable, reduces costs, and lowers the risk of hitting context limits — especially for long-running sessions and tool-heavy agents.

Details

Configurable at the agent, team, and workflow level
Improves latency and reliability by preventing context bloat
Backward compatible: defaults remain unchanged unless you set it

Who this is for: Platform owners and engineering teams running multi-turn or high-volume automations who need tighter cost and performance control.