Quickstart
Get one real run into Neatlogs and use it to understand a production failure.
Choose the shortest path to one useful trace. If you need to instrument an application, paste our prompt into your own coding agent, run the wizard, or use the SDK for your language. If traces already exist in another platform, import them.
You do not need to configure detections, Evals, alerts, or every integration before continuing.
Fastest setup
Instrument with coding agents
Paste one prompt into Claude Code, Codex, Cursor, or any agent that can edit your repository.
Instrument with the Wizard
Run one command that detects your stack and instruments your application.
Import Traces
Bring existing traces from Langfuse, LangSmith, Braintrust, Raindrop, Phoenix, or Laminar.
Instrument manually
Set up with the SDK
Initialize Neatlogs, wrap your client, and verify the first run.
Python SDK
Install Neatlogs and send your first trace from a Python application.
TypeScript SDK
Install the Neatlogs package and send your first trace from a TypeScript or Node.js application.
Integrations
Find setup guides for OpenAI, Anthropic, CrewAI, LangChain, Claude Code, Codex, and more.
The SDK and integration cards link to their canonical pages in the separate top-level sections, which hold the full installation instructions.
Not sure which path fits? Compare the setup methods.
After traces arrive
Verify one complete run
Open Traces and find a run you recognize. Check that the input, final output, important model and tool steps, status, and timing agree with your application. A row in the table is not enough: the trace needs the evidence required to explain its outcome.
Learn what a healthy first trace looks like →
Investigate a real failure
Choose one run with a wrong outcome. Record what happened, what should have happened, and why the difference matters. Attach that context to the evidence, run an investigation, and verify every material conclusion against its citations.
Start with a failure
Follow the complete ten-minute workflow from affected run to reviewable report.
I do not write code
Read evidence, flag problems, and review quality without working with instrumentation.
You do not need a detection, alert, or Eval before the first investigation. Add a detection after you can define a repeatable condition; add an Eval after you can define a repeatable judgment.
You are ready to continue when
- at least one recognizable run appears in the intended project;
- the run exposes enough model, tool, or retrieval evidence to explain its output;
- you can distinguish the observed failure from a theory about its cause;
- you know who should review an investigation before work is handed off.
If the first condition fails, go to Traces are not arriving. If the trace is present but incomplete, revisit the relevant integration or SDK reference.