Telemetry

Agent-first error tracking and observability - no dashboards, just AI

Cased Telemetry is agent-first observability - there’s no dashboard to check, no charts to interpret. Your telemetry data goes directly to AI agents that can query, analyze, and act on it.

Why Agent-First?

Traditional observability tools are built for humans staring at dashboards. You get paged, open a browser, click through charts, and manually investigate.

Today, we can do this better:

• No context switching - Ask your agent “why is the API slow?” and it queries your telemetry directly
• Instant root cause analysis - Agents correlate errors, metrics, and traces automatically
• Automated remediation - Agents can fix issues and open PRs, not just alert you

There’s no pre-baked Cased Telemetry dashboard. The API is the interface, and AI agents are the users.

How It Works

Your Application                         Cased
┌──────────────────┐                    ┌────────────────────────┐
│  SDK             │───────────────────>│                        │
│  (errors/traces) │                    │   Telemetry Storage    │
└──────────────────┘                    │                        │
                                        │          │             │
┌──────────────────┐                    │          ▼             │
│  cased-agent     │───────────────────>│   ┌─────────────┐      │
│  (metrics)       │                    │   │  Query API  │<─────┼─── AI Agents
└──────────────────┘                    │   └─────────────┘      │    or Cased Agents
                                        └────────────────────────┘

Features

• Sentry SDK Compatible - Use the standard sentry-sdk, just change the DSN
• Container Metrics - cased-agent collects CPU, memory, network from Kubernetes
• Distributed Tracing - OpenTelemetry-compatible spans
• LLM Monitoring - Track token usage, costs, latency, and errors across AI integrations
• Performance Analysis - Detect slow spans, N+1 queries, and latency regressions
• Slack Notifications - Get notified when new error types appear
• API & CLI - Query everything via REST API or the Cased CLI

Quick Start

1. Create a Telemetry Project

Create a project in the Cased UI under Connections → Cased Telemetry, or via API:

curl -X POST https://app.cased.com/api/v1/telemetry/projects/ \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "my-app"}'

Either way, you’ll get a DSN like:

https://abc123@telemetry.cased.com/1

2. Configure the SDK

Use the standard Sentry SDK - just point it at your Cased DSN:

uv pip install sentry-sdk

import sentry_sdk

sentry_sdk.init(
    dsn="https://abc123@telemetry.cased.com/1",  # Your Cased DSN
    traces_sample_rate=1.0,
)

For Node.js:

npm install @sentry/node

const Sentry = require("@sentry/node");

Sentry.init({
  dsn: "https://abc123@telemetry.cased.com/1",  // Your Cased DSN
  tracesSampleRate: 1.0,
});

3. (Optional) Deploy cased-agent for Infrastructure Metrics

Deploy the cased-agent DaemonSet to collect container metrics from your Kubernetes cluster:

kubectl apply -f https://raw.githubusercontent.com/cased/cased-agent/main/deploy/manifests/install.yaml

See cased-agent Setup for detailed configuration.

Data Types

Cased Telemetry collects four types of data, each with its own collection method:

Type	Source	What it captures
Errors	Sentry SDK	Exceptions, stack traces, request context
Traces	Sentry SDK	Distributed spans, latency, service dependencies
Metrics	cased-agent	Container CPU, memory, network, restarts
LLM Events	Your code	Token usage, costs, latency per model

Errors (via Sentry SDK)

Captures exceptions with full context: stack traces, request data, user info, and custom tags. Upload source maps to de-minify JavaScript traces.

Traces (via Sentry SDK)

Distributed tracing across services. Analyze latency, find slow spans, detect N+1 queries. See Traces.

Metrics (via cased-agent)

Container-level metrics from Kubernetes: CPU, memory, network I/O, OOM events. Requires cased-agent deployed as a DaemonSet.

LLM Events (via your code)

Track LLM calls with token counts, costs, and latency. Group by session or trace. See LLM Monitoring.

Data Retention

Data Type	Retention
Error Events	90 days
LLM Events	90 days
Metrics	30 days
Traces	14 days

Use with Coding Agents

Once you’ve configured the Cased CLI, your coding agent can query telemetry directly. Just ask:

"Using cased, what errors happened in the last hour?"

"Using cased, why is the checkout API slow?"

"Using cased, show me memory usage for the api-server pods"

No dashboards, no context switching—your agent queries errors, traces, and metrics, correlates them, and can even open PRs with fixes.

Cased Agents

Cased also runs scheduled agent workflows that automatically:

• Detect error spikes and investigate root causes
• Monitor for performance regressions after deploys
• Track LLM cost anomalies
• Create PRs with fixes when issues are found

Slack Notifications

Get notified in Slack when new error types are detected. Configure a Slack channel per project, and Cased will send a notification the first time each unique error is seen—with stack trace, context, and a link to investigate.

Errors are grouped by fingerprint, so you won’t get spammed with duplicates.

Next Steps

• Configure cased-agent for infrastructure metrics
• Monitor LLM usage across your AI applications
• Analyze traces to find bottlenecks and regressions
• Upload source maps for JavaScript de-minification
• Use the CLI to query telemetry from your terminal
• Query the API for programmatic access