Troubleshooting

Common Issues

Data Not Appearing in Dashboard

API key not configured

Symptom: No data appears in dashboard, and you see:

[TCC] No API key found. Set TCC_API_KEY or call configure({ apiKey }).

Solution:Set your API key using environment variables:

TCC_API_KEY=tcc_abc123

Or programmatically:

import { configure } from "@contextcompany/custom";

configure({ apiKey: "tcc_abc123" });

Run never ended

Symptom: Code executes without errors, but no data appears in dashboard.Solution:Ensure you call .end() or .error() on every run:

const r = run();
r.prompt("test");
await r.end(); // Required!

Enable debug mode to verify:

configure({ debug: true });

You should see: [TCC] Payload sent successfully

Steps or tool calls not ended

Symptom: Error when ending run:

[TCC] 2 step(s) not ended. Call .end() on all steps before ending the run.

Solution:All steps and tool calls must be ended before ending the run:

const r = run();
r.prompt("test");

const s = r.step();
s.prompt("...");
s.response("...");
s.end(); // Required!

const tc = r.toolCall("search");
tc.args({ query: "..." });
tc.result({ ... });
tc.end(); // Required!

await r.end();

Missing prompt on run

Symptom: Error when ending run:

[TCC] Run requires a prompt. Call .prompt() before .end()

Solution:Always call .prompt() before ending:

const r = run();
r.prompt("What's the weather?"); // Required!
await r.end();

Network or authentication errors

Symptom: Error in console:

[TCC] Ingestion failed (401): Unauthorized
[TCC] Ingestion failed (403): Invalid API key

Solution:

Verify your API key is correct
Check that the key matches your environment (dev vs production)
Ensure the API key hasn’t been revoked

// Development key (routes to dev environment)
configure({ apiKey: "dev_abc123" });

// Production key (routes to production)
configure({ apiKey: "tcc_abc123" });

Wrong environment

Symptom: Data appears in development dashboard but not production (or vice versa).Solution:API keys automatically route to environments:

Keys starting with dev_ → Development
All other keys → Production

Ensure you’re using the correct API key for your environment.

Debugging

Enable Debug Mode

Debug mode outputs detailed logs to help troubleshoot issues:

import { configure } from "@contextcompany/custom";

configure({ debug: true });

Or via environment variable:

TCC_DEBUG=true

Understanding Debug Output

[TCC] Run created { "runId": "550e8400-..." }
[TCC] Step created { "stepId": "...", "runId": "..." }
[TCC] Run ended { "runId": "..." }
[TCC] Sending payload to https://api.thecontext.company/v1/custom
[TCC] Payload: { "type": "batch", "items": [...] }
[TCC] Payload sent successfully

Verify creation

Look for Run created, Step created, ToolCall created messages.

Verify ending

Look for Run ended, Step ended, ToolCall ended messages.

Verify sending

Look for Sending payload and Payload sent successfully.

Check for errors

Look for Ingestion failed or network errors.

Network Inspection

Inspect the actual HTTP request:

// Enable debug mode
configure({ debug: true });

const r = run();
r.prompt("test");
await r.end();

// Check console for:
// - Request URL
// - Payload structure
// - Response status

Error Messages

Run Errors

Run already ended

You called .end() or .error() multiple times on the same run.

const r = run();
r.prompt("test");
await r.end();
await r.end(); // Error: Run already ended

Solution: Ensure you only end each run once.

Run requires a prompt

You called .end() without first calling .prompt().

const r = run();
await r.end(); // Error: Run requires a prompt

Solution: Always set a prompt before ending.

X step(s) not ended

You created steps but didn’t call .end() on them.

const r = run();
r.prompt("test");
const s = r.step();
s.prompt("...");
s.response("...");
// Missing s.end()!
await r.end(); // Error: 1 step(s) not ended

Solution: Call .end() on all steps before ending the run.

X tool call(s) not ended

You created tool calls but didn’t call .end() on them.Solution: Call .end() on all tool calls before ending the run.

Step Errors

Step already ended

You called .end() or .error() multiple times on the same step.Solution: Ensure you only end each step once.

Step requires a prompt

You called .end() without first calling .prompt().

const s = r.step();
s.response("...");
s.end(); // Error: Step requires a prompt

Solution: Set both prompt and response before ending.

Step requires a response

You called .end() without first calling .response().Solution: Set both prompt and response before ending.

Tool Call Errors

ToolCall already ended

You called .end() or .error() multiple times on the same tool call.Solution: Ensure you only end each tool call once.

ToolCall requires a name

You called .end() without setting a tool name.

const tc = r.toolCall();
tc.args({ ... });
tc.end(); // Error: ToolCall requires a name

Solution: Set the name in the constructor or via .name():

// Option 1: Constructor
const tc = r.toolCall("search");

// Option 2: Method
const tc = r.toolCall();
tc.name("search");

Performance Issues

High Latency

Network latency

Observatory uses async requests with automatic retries. This should not block your application.If you’re experiencing high latency:

Check your network connection to api.thecontext.company
Verify you’re not in a region with high latency to the API
Consider using a custom TCC_URL endpoint closer to your infrastructure

Large payloads

Very large prompts or responses can slow down transmission.Solution:Truncate extremely large content:

const MAX_LENGTH = 50000; // 50KB

const s = r.step();
s.prompt(prompt.slice(0, MAX_LENGTH));
s.response(response.slice(0, MAX_LENGTH));
s.end();

Memory Leaks

Runs not being garbage collected

If you create runs but never end them, they’ll accumulate in memory.Solution:Always end or error runs:

async function safeRun(prompt: string) {
  const r = run();
  r.prompt(prompt);
  
  try {
    const result = await agent.execute(prompt);
    r.response(result);
    await r.end();
    return result;
  } catch (e) {
    await r.error(String(e));
    throw e;
  }
}

Enable auto-flush timeout as a safety net:

configure({ runTimeout: 600000 }); // 10 min

Integration Issues

Next.js

instrumentation.ts not loading

Symptom: No telemetry data captured in Next.js app.Solution:

Ensure instrumentation.ts is in the root of your project (same level as app/ or pages/)
Enable instrumentation in next.config.js:

next.config.js

module.exports = {
  experimental: {
    instrumentationHook: true,
  },
};

Restart your dev server

Server Components issues

Symptom: Errors when using Observatory in React Server Components.Solution:Observatory’s OpenTelemetry integration works automatically in Server Components. For client components, use the widget package.

Vercel

Environment variables not set

Solution:Add TCC_API_KEY to your Vercel project settings:

Go to Project Settings → Environment Variables
Add TCC_API_KEY with your API key
Redeploy

TypeScript

Type errors

Symptom: TypeScript compilation errors.Solution:

Ensure you’re using TypeScript 5.0 or later
Check that types are installed:

pnpm add -D @types/node

Verify your tsconfig.json includes:

{
  "compilerOptions": {
    "moduleResolution": "bundler",
    "module": "ESNext",
    "target": "ES2022"
  }
}

FAQ

How do I test instrumentation locally?

Use a local ingestion endpoint:

configure({ 
  apiKey: "dev_test123",
  url: "http://localhost:8787/custom",
  debug: true 
});

Or just enable debug mode to see payloads in console:

configure({ debug: true });

Can I use Observatory in production?

Yes! Observatory is designed for production use. The SDK:

Uses async requests that don’t block your application
Includes automatic retries for transient failures
Gracefully handles missing API keys (logs warning and continues)
Has minimal performance overhead

What happens if the API is down?

The SDK will:

Retry the request up to 2 times with exponential backoff
Log an error to console after retries are exhausted
Continue running your application normally

Your application will not crash or hang if Observatory is unavailable.

How do I handle timeouts?

Runs have an auto-flush timeout that defaults to 20 minutes:

// Per-run timeout
const r = run({ timeout: 600000 }); // 10 min

// Global timeout
configure({ runTimeout: 600000 });

// Disable timeout
const r = run({ timeout: 0 });

If a run exceeds the timeout, it’s automatically sent with error status.

Can I delete or modify data after sending?

Once data is sent to Observatory, it cannot be modified via the SDK. Contact support if you need to delete or modify historical data.

How do I instrument streaming responses?

For streaming responses, collect the full response before ending the step:

const r = run();
r.prompt("Generate a story");

const s = r.step();
s.prompt(JSON.stringify(messages));

let fullResponse = "";
for await (const chunk of stream) {
  fullResponse += chunk;
  // Stream to user
  yield chunk;
}

s.response(fullResponse);
s.end();

r.response(fullResponse);
await r.end();

How do I handle concurrent runs?

Each run is independent and can be processed concurrently:

const runs = await Promise.all(
  prompts.map(async (prompt) => {
    const r = run();
    r.prompt(prompt);
    const response = await agent.execute(prompt);
    r.response(response);
    await r.end();
    return response;
  })
);

Getting Help

If you’re still experiencing issues:

GitHub Issues

Report bugs or request features

Documentation

Browse the full documentation

Discord Community

Get help from the community

Email Support

Contact the support team

When reporting issues, include:

Debug logs (enable with debug: true)
SDK version
Runtime environment (Node.js version, framework, etc.)
Minimal code reproduction

Get Started

Frameworks

SDKs

Features

Guides

Common Issues

Data Not Appearing in Dashboard

Debugging

Enable Debug Mode

Understanding Debug Output

Network Inspection

Error Messages

Run Errors

Step Errors

Tool Call Errors

Performance Issues

High Latency

Memory Leaks

Integration Issues

Next.js

Vercel

TypeScript

FAQ

Getting Help

GitHub Issues

Documentation

Discord Community

Email Support

Build docs developers (and LLMs) love

Get Started

Frameworks

SDKs

Features

Guides

Documentation Index

​Common Issues

​Data Not Appearing in Dashboard

​Debugging

​Enable Debug Mode

​Understanding Debug Output

​Network Inspection

​Error Messages

​Run Errors

​Step Errors

​Tool Call Errors

​Performance Issues

​High Latency

​Memory Leaks

​Integration Issues

​Next.js

​Vercel

​TypeScript

​FAQ

​Getting Help

GitHub Issues

Documentation

Discord Community

Email Support

Build docs developers (and LLMs) love

Common Issues

Data Not Appearing in Dashboard

Debugging

Enable Debug Mode

Understanding Debug Output

Network Inspection

Error Messages

Run Errors

Step Errors

Tool Call Errors

Performance Issues

High Latency

Memory Leaks

Integration Issues

Next.js

Vercel

TypeScript

FAQ

Getting Help