Termination

Composable stop conditions on Agent.Termination, plus the lighter-weight TextGate for sequential pipelines.

Termination conditions

Combine with & (AND) and | (OR).

var agent = new Agent("researcher")
{
    Termination = new TextMentionTermination("DONE"),
};

// composed
var term   = new MaxMessageTermination(10) | new TextMentionTermination("DONE");
var budget = new TokenUsageTermination(maxTotalTokens: 50_000);

Available: TextMentionTermination, StopMessageTermination, MaxMessageTermination, TokenUsageTermination, and the AndTermination / OrTermination composites produced by the operators.

Text gates

A TextGate stops a sequential pipeline after the agent if its output contains the sentinel text. It is compiled server-side, so there is no worker round-trip:

var checker = new Agent("checker") { Model = "openai/gpt-4o", Gate = new TextGate("STOP") };
var fixer   = new Agent("fixer")   { Model = "openai/gpt-4o" };
var pipeline = checker >> fixer;   // halts after checker if its output contains "STOP"

new TextGate(text, caseSensitive: true) — set caseSensitive: false to match loosely.

Gate vs termination

They solve different problems:

• Termination bounds a single agent's own turn loop — how long this agent keeps going.
• Gate halts a pipeline after this agent, so downstream agents never run.

Turn and token ceilings

MaxTurns on the agent is a hard ceiling independent of Termination, and MaxTokens bounds a single completion. TokenUsageTermination is the cumulative budget across the run.

Reading why a run stopped

AgentResult.FinishReason reports the cause — Stop, Length, ToolCalls, Error, Cancelled, Timeout, Guardrail, or Rejected. See reference/api.md.