Agent execution budget

File: config/system/system-agent.json. Side: server. A missing file is read after being copied from the file of the same name under the JAR's defaults.

Field	Type	Required	Default	Description
timeZone	String	no	"Asia/Shanghai"	Time zone for daily quota settlement; must be a valid ZoneId.
modelCallsPerPlayerDay	Integer	no	100	Per-player daily model call limit, positive integer.
concurrentModels	Integer	no	4	Upper limit of model requests the server executes concurrently, positive integer.
routingPolicy	Object	no	See default file	Router budget, see the table below.
policies	Object	no	{}	Overrides the complete execution budget by workflow ID; this is not a per-player permission setting.

Execution budget object

routingPolicy and every value in policies use the same structure. Except for reactIterations, when overriding one budget item the remaining fields must also be provided in full; a zero model / write quota forbids that operation respectively.

Field	Type	Required	Default	Description
modelCalls	Non-negative integer	yes	—	Maximum number of model calls.
toolCalls	Non-negative integer	yes	—	Maximum number of tool calls.
steps	Positive integer	yes	—	Upper limit of total execution steps.
writeSteps	Non-negative integer	yes	—	Upper limit of write steps.
inputBytes	Positive integer	yes	—	Upper limit of input bytes.
outputTokens	Positive integer	yes	—	Output token budget per request.
totalTokens	Positive integer	yes	—	Token budget for the whole execution.
ttlMillis	Positive integer	yes	—	Total lifetime of the task, in milliseconds.
modelTimeoutMillis	Positive integer	yes	—	Timeout of a single model request, in milliseconds.
reactIterations	Non-negative integer	no	3	Upper limit of the ReAct loop; 0 forbids extra loops.

AI mode limits and per-player authorization are stored in the world SavedData and managed through /system ai; they are not part of this global config file. See AI modes.

Bundled default file

config/system/system-agent.json

{
  "timeZone": "Asia/Shanghai",
  "modelCallsPerPlayerDay": 100,
  "concurrentModels": 4,
  "routingPolicy": {
    "modelCalls": 1,
    "toolCalls": 0,
    "steps": 4,
    "writeSteps": 0,
    "inputBytes": 16384,
    "outputTokens": 512,
    "totalTokens": 32768,
    "ttlMillis": 14500,
    "modelTimeoutMillis": 4000,
    "reactIterations": 0
  },
  "policies": {}
}