See every step your LLM app takes.

TraceLayer turns each request into a call tree — latency, token cost, and input/output for every prompt, retriever, model call, and tool. One line to install. Nothing to rewrite.

Get your API key Explore a live trace ↓

Free plan. No credit card. Key in your hands in a minute.

A tree, not a log.

Nested calls, not flat events.

An example RAG-agent request, running in your browser — no signup. A flat log can't show you that the retriever made the whole call slow.

Click a node or press Enter to expand. Use Tab to move between nodes.

trace rag-agent 5.07 s2,860 tokens$0.0205
prompt Build request 12 ms ·214 tokens . Press Enter to expand.

latency 12 ms

tokens 214

cost $0.0015

input question: "Why did my last invoice go up?"user.email: redacted in the SDK How redaction works →

output Request payload assembled for the retriever.
retriever Vector search 2.87 s ·0 tokens . The slowest step in this trace. Press Enter to expand.

latency 2.87 s — 56.6% of total. The slowest step in this trace.

tokens 0

cost $0.0000

input Query embedding, top-k = 8.

output 8 passages from the billing-docs index.
model call Chat completion 2.06 s ·2,646 tokens . Press Enter to expand.

latency 2.06 s — 40.7% of total.

tokens 2,646

cost $0.0190

input System prompt, the question, and 8 retrieved passages.

output A call to get_invoice, then the drafted answer.
tool Function call 96 ms ·0 tokens . Press Enter to expand.

latency 96 ms

tokens 0

cost $0.0000

input get_invoice(period: "last")

output Invoice line items, returned to the model.
answer Final response 30 ms ·0 tokens . Press Enter to expand.

latency 30 ms

tokens 0

cost $0.0000

input Drafted answer from the model call.

output "Your last invoice went up because two seats were added mid-cycle."

fast / cheap slow / costly

One line to install.

Wrap your code. Traces flow.

One init line wraps your existing code. Traces go out automatically — nothing to restructure, nothing to rewrite.

Read the full quickstart →

Python

install

pip install tracelayer

initialize

import tracelayer

tracelayer.init(api_key="tl_live_...")  # one line — your existing calls are now traced

import tracelayer

# one line — your existing
# calls are now traced
tracelayer.init(
  api_key="tl_live_...")

TypeScript

install

npm install tracelayer

initialize

import { init } from "tracelayer";

init({ apiKey: "tl_live_..." }); // one line — your existing calls are now traced

import { init }
  from "tracelayer";

// one line — your existing
// calls are now traced
init({
  apiKey: "tl_live_...",
});

From one line to a full call tree.

01 Install and initialize
Add the SDK and one init line.
02 Your calls get wrapped
TraceLayer wraps each call in place. It records timing, tokens, cost, and input/output as your code runs.
03 Traces are sent async
The SDK sends each trace to TraceLayer over the REST API.
04 Read the tree in your dashboard
Each request arrives as a call tree. Open any node to see where a request went slow, got expensive, or broke.

Every plan gets every feature — Free included.

You pay for volume. Not for access.

Free 100,000 traces / month $0 No credit card. No expiry.
Growth 2,000,000 traces / month $49/mo
Scale 20,000,000 traces / month $199/mo

See full pricing →

Privacy isn't an upgrade.

Built to stay out of your way.

< 5 ms

per trace · sent async

Trace delivery stays off your request path: your response doesn't wait on it.

Redact in the SDK

Strip sensitive fields before anything leaves your machine. Prompt contents don't have to go over the wire.

Read the full security details →

Read the API before you sign up.

Open to read — no key needed.

POST /v1/tracesSend a trace

GET /v1/traces/:idFetch one trace

GET /v1/tracesList traces

POST /v1/traces

{
  "project":
    "prj_llm_support_bot",
  "name": "rag-agent",
  "spans": [ … ]
}

201 Created

Browse the docs →

Get your key. See your first trace today.

Get your API key

Free plan. No credit card. Key in your hands in a minute.