NewAI Observability

Watchlog AI-Traces

Monitor AI model calls—inputs, outputs, latency, and errors—across services in real time.

Get started free Explore SDKs

30-day trial on paid features. No credit card to start.

AI-Traces dashboard showing inputs, outputs, latency, and error rates
🔒PII-safe redaction
⚡Low overhead
📈Real-time streams

Why use AI-Traces?

  • Capture prompts / inputs and responses / outputs for every AI request.
  • Measure latency, token usage, throughput, and cost by model & route.
  • Detect errors and anomalies with thresholds and smart routing.
  • Analyze usage trends over time and compare models side-by-side.
AI-Traces dashboard preview

Everything you need to observe AI

Prompt & response capture

Record structured inputs/outputs with optional field-level redaction.

Latency & error analytics

Track p50/p90/p99, error codes, and failure hotspots across endpoints.

Tokens & cost by model

Attribute token usage and spend to models, teams, and routes.

Model comparison

Compare error rate, latency, and cost across model providers.

Smart alerts

Thresholds with dedupe; route to Slack, Telegram, or Webhooks.

Team & audit

Invite teammates, share dashboards, and audit configuration changes.

SDKs & Documentation

Node.js SDK

JavaScript / TypeScript

Integrate AI-Traces to capture prompts, responses, latency, tokens, and costs from server routes and workers.

View documentation
Python SDK

Python

Add AI-Traces to Python services to stream inputs/outputs and performance with minimal overhead.

View documentation
Redaction & privacy

Mask PII/PHI before storage with field-level controls.

Metrics & dashboards

Latency, error rate, tokens, cost, model mix.

Alerts & routing

Thresholds, dedupe, Slack/Telegram/Webhook.

Before vs. after AI-Traces

Without
  • No visibility into prompts/responses
  • Latency spikes discovered late
  • Unclear cost and token usage
With Watchlog
  • Full trace of inputs/outputs
  • Real-time latency & error tracking
  • Cost and tokens by model/route

Get started in seconds

Follow our quickstart guide to integrate AI-Traces into your service.

View setup instructions

Frequently asked questions

What does AI-Traces capture?

Prompts/inputs, responses/outputs, latency, token usage, cost, and error details per request.

Which SDKs are available?

Node.js (JavaScript/TypeScript) and Python SDKs with simple initialization.

How do you handle sensitive data?

Enable field-level redaction and masking; TLS enforced in transit.

Can I alert on anomalies?

Yes, configure thresholds and route alerts to Slack, Telegram, or Webhooks with deduplication.

Need help integrating AI-Traces or have questions?

Contact us