A
AmberTrace

Observability for LLM agents

Trace every OpenAI and Anthropic call. Detect silent failures, infinite loops, and wasted tokens. Debug AI agents like you debug code.

Request early accessView on GitHub

AI agents fail silently

Your agent made 47 API calls, burned through 120k tokens, and returned a confidently wrong answer. Your logs show nothing useful. Your users are frustrated.

Traditional observability tools weren't built for this. They track HTTP requests and database queries, not reasoning chains and token economics.

  • Retry loops that waste thousands of tokens
  • Hallucinated tool calls that silently fail
  • Context windows exhausted mid-conversation
  • Model responses that pass validation but miss intent

Trace the reasoning, not just the requests

AmberTrace captures the full execution path of your AI agents. Every prompt, every completion, every tool call, with full token counts and latency metrics.

Full trace visibility

See every LLM call in a timeline view. Understand the sequence of operations, branching decisions, and where things went wrong.

Token economics

Track prompt and completion tokens per call. Identify expensive operations and optimize your prompts based on real usage data.

Failure detection

Automatically flag retry patterns, unusually long chains, and error responses. Get alerted before your users notice.

Multi-provider support

Works with OpenAI and Anthropic out of the box. Same dashboard, same traces, regardless of which model you're calling.

Three lines to instrument your agent

1

Install the SDK

pip install ambertrace or npm install ambertrace

2

Initialize with your API key

One line at the start of your application. No code changes to your LLM calls.

3

View traces in real-time

Every call appears in your dashboard immediately. Filter by model, status, or duration.

Works with your existing code

AmberTrace patches OpenAI and Anthropic clients automatically. No wrapper functions, no decorators, no manual instrumentation.

Python
main.py
import ambertrace
from openai import OpenAI

ambertrace.init(api_key="at_...")

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Analyze this data"}]
)

# Every LLM call is now traced automatically
TypeScript
index.ts
import { init } from 'ambertrace';
import OpenAI from 'openai';

init({ apiKey: 'at_...' });

const client = new OpenAI();
const response = await client.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'Analyze this data' }]
});

// Every LLM call is now traced automatically

Built for teams shipping AI to production

AI Engineers

Debug agent behavior without adding print statements everywhere. Understand why your agent made specific decisions and where it went off track.

Technical Founders

Ship AI features faster with confidence. Know exactly how much each user session costs and identify optimization opportunities before they become problems.

Platform Teams

Give your AI teams the same observability infrastructure they expect for traditional services. Standardize on a single tool for all LLM monitoring.

Stop debugging AI agents blindly

AmberTrace is currently in early access. Request an invite and start tracing your LLM calls today.

Request early access

Open source SDKs. Self-host or use our cloud.