Lindy Observability

Overview

Implement comprehensive observability for Lindy AI integrations.

Prerequisites

• Production Lindy integration
• Observability stack (Datadog, New Relic, Prometheus, etc.)
• Log aggregation system

Instructions

Step 1: Structured Logging

// lib/logger.ts
import pino from 'pino';

export const logger = pino({
  level: process.env.LOG_LEVEL || 'info',
  formatters: {
    level: (label) => ({ level: label }),
  },
  base: {
    service: 'lindy-integration',
    environment: process.env.NODE_ENV,
  },
});

// Lindy-specific logger
export function lindyLogger(operation: string) {
  return logger.child({ component: 'lindy', operation });
}

Step 2: Instrumented Client

// lib/instrumented-lindy.ts
import { Lindy } from '@lindy-ai/sdk';
import { lindyLogger } from './logger';
import { metrics } from './metrics';
import { tracer } from './tracer';

export class InstrumentedLindy {
  private lindy: Lindy;

  constructor() {
    this.lindy = new Lindy({ apiKey: process.env.LINDY_API_KEY });
  }

  async runAgent(agentId: string, input: string) {
    const log = lindyLogger('runAgent');
    const span = tracer.startSpan('lindy.agent.run');

    const startTime = Date.now();

    try {
      span.setAttributes({
        'lindy.agent_id': agentId,
        'lindy.input_length': input.length,
      });

      log.info({ agentId, inputLength: input.length }, 'Starting agent run');

      const result = await this.lindy.agents.run(agentId, { input });

      const duration = Date.now() - startTime;

      // Record metrics
      metrics.histogram('lindy.agent.duration', duration, { agentId });
      metrics.counter('lindy.agent.success', 1, { agentId });

      // Log success
      log.info({
        agentId,
        duration,
        outputLength: result.output.length,
      }, 'Agent run completed');

      span.setAttributes({
        'lindy.duration_ms': duration,
        'lindy.output_length': result.output.length,
        'lindy.status': 'success',
      });

      return result;
    } catch (error: any) {
      const duration = Date.now() - startTime;

      // Record error metrics
      metrics.counter('lindy.agent.error', 1, {
        agentId,
        errorCode: error.code,
      });

      // Log error
      log.error({
        agentId,
        duration,
        error: error.message,
        errorCode: error.code,
      }, 'Agent run failed');

      span.setAttributes({
        'lindy.status': 'error',
        'lindy.error': error.message,
      });
      span.recordException(error);

      throw error;
    } finally {
      span.end();
    }
  }
}

Step 3: Metrics Collection

// lib/metrics.ts
import { Counter, Histogram, Registry } from 'prom-client';

const registry = new Registry();

export const metrics = {
  agentDuration: new Histogram({
    name: 'lindy_agent_duration_ms',
    help: 'Duration of Lindy agent runs in milliseconds',
    labelNames: ['agent_id', 'status'],
    buckets: [100, 500, 1000, 2000, 5000, 10000, 30000],
    registers: [registry],
  }),

  agentRuns: new Counter({
    name: 'lindy_agent_runs_total',
    help: 'Total number of Lindy agent runs',
    labelNames: ['agent_id', 'status'],
    registers: [registry],
  }),

  apiCalls: new Counter({
    name: 'lindy_api_calls_total',
    help: 'Total Lindy API calls',
    labelNames: ['endpoint', 'status'],
    registers: [registry],
  }),

  // Helper methods
  histogram(name: string, value: number, labels: Record<string, string>) {
    const metric = registry.getSingleMetric(name) as Histogram;
    metric?.observe(labels, value);
  },

  counter(name: string, value: number, labels: Record<string, string>) {
    const metric = registry.getSingleMetric(name) as Counter;
    metric?.inc(labels, value);
  },
};

// Metrics endpoint
export function getMetrics(): Promise<string> {
  return registry.metrics();
}

Step 4: Distributed Tracing

// lib/tracer.ts
import { trace, SpanStatusCode } from '@opentelemetry/api';
import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
import { SimpleSpanProcessor } from '@opentelemetry/sdk-trace-base';
import { OTLPTraceExporter } from '@opentelemetry/exporter-trace-otlp-http';

const provider = new NodeTracerProvider();

provider.addSpanProcessor(
  new SimpleSpanProcessor(
    new OTLPTraceExporter({
      url: process.env.OTEL_EXPORTER_OTLP_ENDPOINT,
    })
  )
);

provider.register();

export const tracer = trace.getTracer('lindy-integration');

Step 5: Dashboard Configuration

# grafana/dashboards/lindy.json
{
  "title": "Lindy AI Monitoring",
  "panels": [
    {
      "title": "Agent Runs per Minute",
      "type": "graph",
      "targets": [
        {
          "expr": "rate(lindy_agent_runs_total[1m])",
          "legendFormat": "{{agent_id}}"
        }
      ]
    },
    {
      "title": "P95 Latency",
      "type": "stat",
      "targets": [
        {
          "expr": "histogram_quantile(0.95, rate(lindy_agent_duration_ms_bucket[5m]))"
        }
      ]
    },
    {
      "title": "Error Rate",
      "type": "graph",
      "targets": [
        {
          "expr": "rate(lindy_agent_runs_total{status='error'}[5m]) / rate(lindy_agent_runs_total[5m])"
        }
      ]
    }
  ]
}

Output

• Structured logging
• Prometheus metrics
• Distributed tracing
• Grafana dashboards
• Alerting rules

Error Handling

Issue	Cause	Solution
Missing traces	OTEL not configured	Set OTEL endpoint
Metrics not visible	Wrong labels	Check label names
Logs not searchable	Missing context	Add structured fields

Examples

Alert Configuration

# alerts/lindy.yml
groups:
  - name: lindy
    rules:
      - alert: LindyHighErrorRate
        expr: rate(lindy_agent_runs_total{status="error"}[5m]) > 0.1
        for: 5m
        labels:
          severity: critical
        annotations:
          summary: "High Lindy error rate"

      - alert: LindyHighLatency
        expr: histogram_quantile(0.95, rate(lindy_agent_duration_ms_bucket[5m])) > 10000
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "Lindy P95 latency above 10s"

Resources

• OpenTelemetry
• Prometheus
• Grafana

Next Steps

Proceed to lindy-incident-runbook for incident response.