FluxyChatAI Agents
Agent lifecycle callbacks
Systematic onStart, onStepStart, onToolExecution, and onEnd hooks for generate, stream, and agent loops.
@fluxy-chat/agent exposes lifecycle hooks aligned with Vercel AI SDK 7 telemetry phases — use them for logging, billing, guardrails, and OpenTelemetry export.
Callback phases
| Phase | When |
|---|---|
onStart | Before the first model step |
onStepStart | Before each agent step |
onLanguageModelCallStart / onLanguageModelCallEnd | Around each LLM request |
onToolExecutionStart / onToolExecutionEnd | Around each tool call |
onStepEnd | After each step completes |
onEnd | When the run finishes |
Agent loop example
import { runAgentLoop } from "@fluxy-chat/agent";
await runAgentLoop({
model,
tools,
messages,
onStart: ({ callId, maxSteps }) => {
console.log("run", callId, "maxSteps", maxSteps);
},
onToolExecutionStart: ({ toolCall }) => {
console.log("tool", toolCall.name);
},
onEnd: ({ text, usage }) => {
console.log("tokens", usage.totalTokens);
},
});SDK telemetry bridge
Wire the same events to OTLP or console export:
import {
createTelemetryManager,
createOtlpTelemetryIntegration,
} from "@fluxy-chat/sdk";
const telemetry = createTelemetryManager();
telemetry.register(
createOtlpTelemetryIntegration({
endpoint: process.env.OTEL_EXPORTER_OTLP_TRACES_ENDPOINT!,
}),
);See Observability (OTLP).