Termination + TextGate

Termination conditions decide when a multi-turn / multi-agent loop should stop. Pass one to termination:. They compose with .and() / .or() (or the variadic AndCondition / OrCondition).

import { TextMention, MaxMessage, TokenUsageCondition, StopMessage } from '@io-orkes/conductor-javascript/agents';

const agent = new Agent({
  name: 'debate',
  model,
  agents: [a, b],
  strategy: 'round_robin',
  termination: new TextMention('TERMINATE')               // stop when output mentions text
    .or(new MaxMessage(10))                                // …or after 10 messages
    .or(new TokenUsageCondition({ maxTotalTokens: 50000 })),
});

Available conditions: TextMention(text, caseSensitive?), StopMessage(stopMessage), MaxMessage(maxMessages), TokenUsageCondition({ maxTotalTokens?, maxPromptTokens?, maxCompletionTokens? }), and the composites AndCondition(...) / OrCondition(...).

TextGate and gate() gate transitions (e.g. on gate:):

import { TextGate } from '@io-orkes/conductor-javascript/agents';
new Agent({ name: 'a', model, gate: new TextGate({ text: 'APPROVED', caseSensitive: false }) });

Bounding cost and stopping safely

Set a meaningful maxTurns in addition to any termination condition — it's the backstop when a text/token condition never triggers. Stop a live execution through AgentHandle.stop() / AgentClient.stop(): make the call safe to repeat, and don't assume an in-flight external tool call is reversible — if a tool has already started, stopping the agent doesn't undo its side effect. Design compensating work where a stopped execution can leave external state partially applied.

Next steps

Continue with multi-agent and agent client control.