Your own receipts

Where does your agent lose you?

Point Emulo at your agent logs. It finds the messages where you sent the same ask again, wrote that you had already said it, rephrased the same request, or opened with a correction.

The files stay on this machine. This page reads the folder in your browser, runs the checks here, and makes no network request. Closing the tab clears the result.

Emulo scanready
01

Choose your log folder

Pick the folder itself. Emulo searches its subfolders for JSONL sessions.

Copy one, then paste it into the folder dialog. Both folders are hidden by default, so typing the path is faster than browsing for it.

Chrome and Edge use direct folder access. The browser fallback works without it.

02

Reading your side of the conversation

Finding JSONL session files…

0 / 0 files0.0s
03

Your scan

These are text matches on your own messages. They cannot tell why you repeated something or whether the agent forgot anything; each finding shows its receipts so you can judge.

These checks only read messages you typed. They cannot see model quality, token cost, tool calls, or whether a correction was justified.

The obvious next move, and whether it works

Most people answer a scan like this by writing the agent a rules file. I tested my own rules file first: three tasks, the same model, once with the rules loaded and once without, one run each. Both runs found the same four real defects. With the rules it took 3.2 times as long, and it brought in two reviewers, one of whom caught a defect that run had shipped. On raw defect discovery, the rules made no measurable difference.

Read what the test found.

If you want that harness pointed at your repo and your rules, write to founder@viberaven.dev and say what your agent keeps getting wrong. Ohad reads them.

!

Nothing to scan yet