Troubleshooting
The five things that actually go wrong, and what fixes them.
Windows says the file isn’t safe
The blue full-screen warning on first run means the build is not signed with a code-signing certificate yet — not that anything is wrong with the file. Open the extra details the warning offers and it reveals the button that runs the installer anyway. Why Windows warns about this file explains it in full and shows how to check the download against the SHA-256 we publish.
It doesn’t hear the call
The level meter under Recording stays flat while people are talking.
- Check the permission on the Permissions tab. If it says Denied, press the button there — it opens the right settings panel.
- jhint hears what your computer plays. If the call’s audio goes to a headset that the computer doesn’t route through itself, there is nothing to capture.
After the permission is granted, macOS asks to quit and reopen jhint. Until you do, the app still can’t hear anything.
There is no permission to check here: jhint captures what your computer plays without asking Windows for anything. If the meter is flat, the audio is going somewhere jhint isn’t — a headset the system doesn’t route through itself, or an app playing through a separate device.
Words come out wrong
Name the speech language in the New session dialog instead of leaving it on Auto — that alone fixes most of it. If the other side is quiet or distant, their audio is the limit, not the model. A larger model on the Speech recognition tab helps with accents and poor audio, at the cost of a slower first start.
The answer takes too long
With the built-in model: the first session after launch spends a few seconds starting the engine. If your computer is short on memory, a 4B model will behave much better than an 8B one.
With your own provider: it is the network and the provider. Some reasoning models take tens of seconds by design — pick a faster model for live calls.
Either way, Stop cancels an answer you no longer need.
The model won’t download
The setup window and the LLM Provider tab both show the state and a Retry button. A download that failed resumes from where it stopped rather than starting over. If it fails repeatedly, a corporate network is the usual reason.
It won’t start a session
- No licence — the key isn’t activated yet; see License.
- The model is still downloading — the window says which one and how far along it is.
- The speech model is preparing — the first run prepares it for your hardware, which takes a few minutes and happens once.
Still stuck
Open General, read the Diagnostics block, and send it to support@jhint.ai. It says the version, the permission state, the provider and model, and where data is stored — enough for us to answer without a round of questions.