Tool Call Cap
maxToolCallsPerRound limits the number of tool calls the agent can make in a single inference round. Calls beyond the cap are silently dropped before execution, preventing runaway tool use on a single turn.
Config
yaml
agents:
my-agent:
tools:
allow: [search, write_file, run_code]
maxToolCallsPerRound: 8 # default: 8Fields
| Field | Type | Default | Description |
|---|---|---|---|
maxToolCallsPerRound | integer | 8 | Max tool calls executed per inference round |
Behaviour
If the model returns 12 tool calls and maxToolCallsPerRound is 8, the first 8 calls execute and the remaining 4 are discarded. The agent continues to the next round normally.
💡Lower this value (e.g. 3–4) for agents that should stay focused on a single action per turn. Raise it for parallel research or batch-processing agents.