Lag and instability issues

[Gamer-Hacker123]

Currently, I have used Atomic Agent in Windows on Powershell, and it has some instability issues. For example, when I add a custom endpoint, it freezes and doesn't work until I manually terminate that instance of Powershell.

[sosidudku1]

Thanks for reporting. To fix this we need a bit more detail, the description does not yet say enough to reproduce it:

1. Which endpoint were you adding, and where: the external llama.cpp URL in the LLM tab, or a cloud provider with a custom base URL in the setup wizard?
2. Which server was running behind it (KoboldCpp from Feature request: Please add support to other LLM runners #66, plain llama.cpp, something else) and on which port?
3. At which exact step does it freeze: right after you press Enter on the URL, during the connection check, or later when you send the first message?
4. Does anything appear on screen before it stops responding (an error line, a spinner, a status message)?
5. Which version are you on, and does Ctrl+C do anything, or is closing the window the only way out?

A screenshot of the frozen screen would help a lot too, like the one you attached in #66.

[Gamer-Hacker123]

@sosidudku1 Thanks for responding on the issue report.

1. I was adding it in the external llama.cpp .
2. It was KoboldCpp on port 5001 .
3. Immediately after I hit Enter on the URL
4. No, nothing shows on the screen, the agent runs perfectly normal, until I enter the URL of KoboldCpp and hit Enter .
5. The latest version according to 07/07/2026 (DD/MM/YYYY), installed using the command irm https://atomicagent.io/install.ps1 | iex on Powershell
6. But honestly, Immediately after I hit Enter on the keyboard, it just freezes and deosn't respond whatever I do or whatever key I press. And the only way to terminate it, is to terminate that instance/window of Atomic Agent . Even if I hit Ctrl + C or Esc it doesn't respond at all.

Important detail

the window freezes completely when the endpoint is exactly: http://localhost:5001(without v1) but, if I add the v1 in the end of the endpoint, it freezes just for a while then I can edit the endpoint again.

In other words: the localhost:5001 without the v1 freezes completely, but when I add v1 in the end it freezes just for seconds then it can be editable.

7. Here's the screenshot, but the screenshot doesn't show much details actually, it just shows the TUI.

Hope you get the needed information, and hope you can fix the issues so this project grows 💖

[sosidudku1]

This is a great writeup, thank you, the /v1 detail cracked it. What you see as a short freeze is our URL probe running up to five retries with a three second timeout each, synchronously, so the UI blocks for up to ~20 seconds. The permanent freeze without /v1 looks like KoboldCpp answering 200 to our llama.cpp-specific /health check, which makes the probe falsely pass and a later call hang.

Fix planned: make the probe non-blocking and cancellable, cut the retries for manual URL entry, and probe /v1/models as a fallback so KoboldCpp and other OpenAI-compatible runners pass honestly (that also covers #66). Keeping both issues open until it ships.

[Gamer-Hacker123]

@sosidudku1 Thanks again.

But just to clear things up: When I do add the v1 in the end of the endpoint (localhost:5001/v1), and hit Enter, it doesn't freeze for too long, it freezes just for about 1 to 2.5 seconds, not ~20 seconds or more.

And again, thanks for your understanding, and I hope this project nails the interest and the love it deserves .