AI Error Monitoring & Automatic Investigations — DeepTracer
Guardian Mode · Always watching

You built with AI. Now your app has
its own.

DeepTracer is an AI agent for production apps. It monitors your logs 24/7, spots problems before users notice, and runs investigations automatically — telling you exactly what broke and how to fix it.

Works with
Next.js Vercel Lovable Cursor v0 Node.js
deeptracer · agent · guardian-mode
ACTIVE
1,284,719 events processed
dt
Watching my-app.vercel.app
Status
Healthy
Investigations
247 total
LLM cost
$4.21 today
Activity Feed
Investigation #247 NEW
Stripe webhook timing out on checkout
Root cause: STRIPE_SECRET_KEY set in dev but missing in Vercel production env. 3 users affected, $240 in failed checkouts.
3 users affected $240 lost Fix ready
View full investigation →
Ambient Monitoring · Guardian Mode

Your agent never sleeps.

Every 5 minutes, your agent runs a full health check — error rates, response times, LLM costs. The moment something looks off, it investigates automatically. By the time you wake up, the root cause is already found.

Live health monitoring · last night · my-app.vercel.app
12:00 AM
Healthy
847 checks · all clear
3:00 AM
Healthy
Error rate +340%
3:47 AM
Alert
Auto-investigation →
3:52 AM
Investigating
Root cause · 43s
3:53 AM
Fix ready
Notification waiting ▼
8:00 AM
You wake up ☀
5-minute health cycles
288 checks per day, automatically. Your agent scans error rates, response times, LLM costs, and traffic patterns — looking for anything that doesn’t look right.
Auto-trigger investigations
The moment a threshold is crossed, your agent starts reading logs, tracing requests, and building the investigation. No manual trigger, no waiting until morning.
Fix-first alerts
Your Slack notification doesn’t just say something broke. It includes what broke, why it broke, and how to fix it — before you’ve even opened your laptop.
What arrives in Slack at 3:53 AM
# deeptracer-alerts
dt
DeepTracer APP 3:53 AM

🔔 Investigation complete — root cause found in 43 seconds.

Stripe webhook timing out on /api/checkout
Root cause: STRIPE_SECRET_KEY is set in .env.local but missing from Vercel production environment variables. Added in dev, never promoted to production.

Impact: 3 users failed checkout since 3:47 AM · $240 in failed transactions
Fix: Add STRIPE_SECRET_KEY to Vercel dashboard → Settings → Environment Variables → redeploy.
View full investigation →
AI Chat · Ask anything

Ask your agent anything.

Can’t figure out why something broke? Just ask. Your agent reads your logs, finds the pattern, and explains it in plain English — no dashboards to learn, no query language to write.

AI Chat · my-app.vercel.app
Guardian Mode
You · just now
why did my OpenAI bill spike yesterday?
dt
DeepTracer AI INVESTIGATION responded in 6s

I analyzed your LLM usage for the past 24 hours. Your costs jumped from $6.79/day to $47.18/day — a 595% increase. Here’s what happened:

Root Cause

Your system prompt grew from ~120 tokens to ~3,800 tokens after the 2:14pm deploy (commit a3f9b12 “add full docs to system context”). Every GPT-4 Turbo request now sends 3,500+ tokens before the user says a word — multiplied across 1,247 API calls yesterday.

Evidence
SPIKE 2:14pm – now  GPT-4 Turbo · 1,247 calls · 5.2M tokens · $47.18
NORMAL prev 7 days avg  GPT-4 Turbo · 891 calls · 754K tokens · $6.79
DEPLOY 2:14pm  commit a3f9b12 — system prompt: 121 tokens3,847 tokens
Most Expensive Endpoints
/api/chat $0.038/req avg 1,247 calls · $47.38 today
/api/summarize $0.004/req avg 89 calls · $0.36 today
Suggested Fix
const systemPrompt = fullDocumentation; // 3,847 tokens
+ const systemPrompt = shortSummary; // 120 tokens, cached
Estimated Savings
$35–40 saved per day $40.39 overspend yesterday 595% cost increase
Ask anything about your app…
Try asking:
“Why did checkout fail at 3am?” “What’s making my app slow?” “What changed in the last deploy?” “Any errors I haven’t noticed?”
LLM Monitoring · Included in Pro

Your AI features
have hidden costs.

Every GPT-4 call, every Claude message — your app generates AI costs you can’t see in Vercel or your error tracker. DeepTracer tracks every LLM request, shows your cost per user, and alerts you before your invoice is a surprise.

  • Works with OpenAI, Anthropic, Google Gemini, Mistral, and more
  • Cost-per-user breakdown by model, endpoint, and time period
  • Spike alerts — get a Slack message the moment costs exceed your threshold
🔥

Helicone charges $79/mo for LLM monitoring alone. DeepTracer includes it in Pro — alongside error tracking, ambient monitoring, and AI investigations — for $19/mo.

LLM Cost · last 7 days
● $6.79 today
7-day total
$84.57
Cost / user
$0.038
Peak (Sun)
$47.18 (fixed)
Daily LLM spend
Mon
Tue
Wed
Thu
Fri
Sat
Deploy 2:14pm
Sun
gpt-4o
847 calls $38.40
claude-haiku-4-5
1,247 calls $6.12
gpt-3.5-turbo
89 calls $0.09
On track for ~$210/mo at current volume · set a threshold to get alerted
Setup · 5 minutes

Five lines of code. Done.

No dashboards to configure. No alert rules to write. Install the SDK, add five lines, and your AI agent starts watching immediately.

Step 1
$ npm i @deeptracer/nextjs
Step 2
instrumentation.ts
Next.js
Node.js
Express
1 2 3 4 5 6 7
import { DeepTracer } from '@deeptracer/nextjs'; export function register() { DeepTracer.init({ apiKey: process.env.DEEPTRACER_KEY, }); }
1 2 3 4 5 6 7 8
import { DeepTracer } from '@deeptracer/node'; // Call before anything else in your entry file DeepTracer.init({ apiKey: process.env.DEEPTRACER_KEY, service: 'my-api', // optional label });
1 2 3 4 5 6 7 8
import { DeepTracer } from '@deeptracer/node'; // Must init before express() DeepTracer.init({ apiKey: process.env.DEEPTRACER_KEY }); import express from 'express'; const app = express(); // your routes...
Agent connected · my-app.vercel.app
TypeScript
✓ Errors tracked ✓ Ambient monitoring ✓ AI investigations ✓ LLM cost tracking
Read the full setup guide →
FAQ

Straight answers.

The questions we get most from developers before they add that first line of code.

How does the AI investigation actually work?

When your agent detects an anomaly — or you ask a question — it pulls the relevant log entries, error traces, and recent deploy history. That context is sent to Claude (our AI) with a structured prompt asking for root cause, evidence, and a suggested fix.

The whole process typically takes 20–60 seconds. Your code and logs are never used to train AI models, and we pass only the minimum relevant context per investigation.

Will the SDK slow down my app?

No. The SDK is async and non-blocking — events are batched and flushed in the background on a 2-second cycle. If our ingestion service is unreachable, events drop silently without affecting your app.

@deeptracer/core has zero dependencies. Typical overhead is <1ms per request.

What’s the difference between Reactive and Guardian mode?

Reactive (Free): your agent only investigates when you manually trigger it — like a doctor you visit. You get 3 investigations per month.

Guardian (Pro, $19/mo): your agent watches 24/7, runs health checks every 5 minutes, auto-investigates when thresholds are crossed, and sends Slack alerts before your users notice. Most people upgrade during a late-night debugging session after hitting the 3-investigation limit.

Do you train AI on my logs?

No. Never. Your logs are used only to generate investigation responses for you — they’re not retained by our AI provider beyond the API call, and we never use customer data to train models.

All data is encrypted in transit (TLS 1.3) and at rest. You can delete your data at any time from the dashboard.

What counts as an event?

Every log line, error, trace span, or LLM call your app sends counts as one event. A typical Next.js app with moderate traffic generates 10K–50K events/month. The free tier (25K/mo) covers most side projects.

If you go over, we never cut you off. Overages are billed at $0.50 per 100K events and you get an email at 80% and 100% usage.

Can I run DeepTracer alongside Sentry?

Yes, they don’t conflict. Many users run both for a few weeks while getting comfortable, then drop Sentry once they realise they don’t need it.

DeepTracer tracks errors, LLM costs, logs, and traces in one place. Sentry only does errors. When you’re ready to switch fully, you can disable Sentry in one line.

More questions? Email us — we reply within a few hours.

Free to start · Guardian mode for $19/mo

Your app built with AI
deserves an AI watching it.

Add 5 lines of code. Your agent starts watching immediately. Keep it on the free tier or unlock Guardian mode when you’re ready.

Free tier available · No credit card required · $19/mo to unlock Guardian mode