Agent setup
Let your coding agent run the audit.
Paste this into Claude Code, Cursor, Codex or any agent that can read a web page and run a shell command, from the repo that makes your LLM calls.
Read https://sundr.ai/setup.md and follow it to set up a Sundr Savings Audit for this repo.
The agent works with the credentials already on your machine. Nothing is uploaded until you approve the exact file.
- Doesn't change your code
- Never prints or uploads API keys
- Shows you the file, row count and label mix before asking to upload
- Finds the judgment callsLLM calls that return one of a fixed set of labels: classify, moderate, route, grade. You pick which one to audit.
- Finds the logsLangfuse, Helicone, LangSmith, Braintrust, your own database or log files, using the access you already have.
- Writes one export fileEach call's input and the label your LLM returned, plus monthly volume and cost per call worked out from your code and logs.
- Asks before uploadingYou see the file, a few example rows and where the report goes. It uploads only on your yes.
- You confirm by emailThe audit starts when you click the link, and the report arrives by email: the share of calls Sundr could answer at each error budget, certified on your data.