Setup steps, answers and fixes for Private Local Chat. This page is plain HTML: it needs no model and no script, and once the app is saved on your device it opens without a connection.
Open the app. Go to https://privatelocal.chat/ in a current browser. There is nothing to install and no account to create.
Write a message and send it. With on-device chat, the first message sets everything up. The app picks a model that suits your device (the smaller Qwen3.5-2B, about 1.20 GB, on phones and tablets and small devices; a capable computer may get the larger Qwen3.5-9B, about 5.06 GB), shows its download size, downloads it, loads it and then answers. Keep the page open while it works. A progress card shows the stage, the size and the time left, and you can cancel.
Prefer an outside provider? In the app's model settings, set Where replies run to Connected provider, choose an endpoint and enter the API key from that provider. Messages in that mode go to the provider you chose; see what happens to my messages.
Check readiness. The model settings show three separate facts: downloaded to this device, loaded in memory, and ready for offline use. See offline readiness before you rely on it without a connection.
Keep your work safe. Conversations live in this browser only. Use Storage & backups in the app to back up chats, and read restoring a backup.
Private Local Chat lets you use AI in your web browser. Run a supported model on your own device or choose to connect an outside provider. Use it to draft, ask questions, explore ideas, and work through information.
The app is free to use. On-device chat works without a Private Local Chat account or API key. Optional outside providers may require their own accounts and may charge for service.
Why does it download a model?
Model files let your device generate replies locally. The first download can be substantial. Saved files can be reused while they remain in browser storage. Check the displayed download size and device guidance during setup.
On a supported device, on-device text chat can work offline after the app and model are saved and ready. Check the readiness indicators and try it without a connection before relying on it while traveling. Connected providers and some voice features require network access.
In on-device text mode, the model processes conversation content on your device. When you choose a connected provider, the messages, instructions, document excerpts, and other context included in a request go to that provider. Its data practices apply to those requests.
Voice uses the speech services available in your browser. Dictation and voice conversation may send audio to the browser's speech service. Speech processing and the location where text replies are generated are separate capabilities.
Compatibility and performance depend on the device, browser, available memory, and selected model. Follow the setup guidance and consult the dated known-issues section. Include your device, browser, and model when reporting setup or performance problems.
Saved conversations are kept in this browser's storage. Each browser and device has its own storage. Back up important work before clearing site data or moving devices. Browser storage removal or loss of the device can affect what remains available.
App lock provides a passphrase screen that helps prevent casual access through the app. Saved conversation and document data remain readable to someone with access to the browser's stored files. Encrypted backups provide a separately protected copy of the included data.
Free memory unloads the model from working memory and keeps its downloaded files. Remove download deletes the saved model files, which can be downloaded again later. Back up important conversations before clearing all site data.
Review generated work and check facts that matter. Models differ in capability and can make mistakes, misunderstand a question, or invent details.
How can I recommend Private Local Chat?
Open Share and choose your device's sharing option, copy the link, or display the QR code. The public address is https://privatelocal.chat/. Each person uses their own device and browser storage.
Open Transfer settings, review the included fields, and choose whether to include your API token. An included token is encrypted with the password you enter. Create the link or QR code and communicate the password separately. The other listed settings remain readable to the recipient.
Import the ordinary settings and choose Continue without API token. You can retry the password later, enter a token manually, or create a new transfer from a device that still has the original token. A provider requiring authentication needs a usable token before it can answer.
How do I move my conversations?
Open Transfer chats and backups, review the file's included data and protection, and export it. Import the file on the receiving device and review the Merge or Replace choice. Transfer provider configuration separately through Transfer settings.
Use your browser's Install or Add to Home Screen option where available. Check app and model readiness before relying on offline use.
Share Private Local Chat
Troubleshooting
Each problem below has its own link, so you can share the exact fix. Do not paste API keys or private conversations into a public report.
A model download stalls or keeps restarting
Downloads are split into files that are checked as they arrive, and finished files are kept. If no data arrives for two minutes, the app stops that transfer and restarts it by itself, reusing the files already saved. The progress card then says “The download stalled. Restarting it now; nothing already downloaded is lost.” If the connection drops, the download waits and continues when you are back online. A closed browser cannot keep downloading; it continues when you reopen the page.
Keep the page open and the device awake, and check your connection. The model files come from assets.privatelocal.chat, so a firewall, VPN or content blocker that blocks that address stops the download.
Check free space on the device. The smaller model is about 1.20 GB and the larger one about 5.06 GB. When storage is full, saving files fails again and again.
Choose the smaller model: use “Use the smaller model instead” on the progress card where it is offered, or pick Qwen3.5-2B in the model settings.
Open Advanced model controls in the model settings and choose Verify / repair. It removes only damaged files and downloads them again.
Still stuck? Choose Cancel download, reload the page and start again; files that are already complete are kept. Then report it with your device, browser and model (see contact).
A model will not load
Loading puts the downloaded model into memory. Common causes and what to do:
Not enough memory. The larger model needs about 6 GB and up to about 9 GB at its peak. Close other tabs and heavy apps, use Free memory if another model is loaded, or choose the smaller Qwen3.5-2B model.
No usable graphics card (WebGPU). The model settings say what the browser reported and what to try, such as updating the browser (Safari 26 or newer, or a current Chrome) or the graphics driver. When no GPU is usable, the smaller model can run on the processor, which is slower. That is not offered on iPhone or iPad, and the larger model needs a GPU.
Damaged or incomplete files. Open Advanced model controls and choose Verify / repair.
The model engine stopped. In Advanced model controls choose Restart model engine, then Load model.
If it still fails, choose Copy device details in Advanced model controls. It copies the device, browser and model state, with no conversation text; read it before you share it. Include it, and the model you chose, in a report (see contact). Related: will it work on my device?
The page reloads or restarts by itself
A new version was published. The app clears its saved copy and reloads once. If that already happened, a banner says “A newer version of PrivateLocal.Chat is available” with Update now. If the app cannot start, a banner says it “did not start properly” with Update the app.
The browser replaced the page. Phone browsers can replace a page that uses a lot of memory. This has been seen on an iPhone while loading a model and during long replies; see the known issue.
Your saved conversations, unsent drafts (kept per conversation) and downloaded model files stay saved. To reduce the chance of it happening again, close other tabs and apps, choose the smaller model, or set Where replies run to Connected provider or use a computer. If it keeps happening, report the device, browser, model and what you were doing (see contact).
Checking that it will work offline
The model settings list three separate facts: Downloaded to this device, Loaded in memory, ready to answer and Ready for offline use (app and model saved). All three matter.
Set up on-device chat while online, so the model is downloaded.
Reload the app once while online. The app itself is saved for offline use, and the settings say “Reload the app once while online so the app itself is saved for offline use too” when that is still needed. After an update they say “An updated app is ready. Reload while online before the offline test.”
When the settings say “Everything needed is saved on this device. You can use it without a connection.”, turn on airplane mode, reopen the app and send a message. Do this test before you travel.
Browsers can clear saved data when the device is low on space. In the app, Keep models on this device (Advanced model controls) and Keep my data (Storage & backups) ask the browser to protect them. The browser may decline.
Connected providers, downloading a model and some voice features need a connection. Once the app is saved, these Help pages open offline too.
Freeing memory or removing a model
Free memory unloads the model from working memory and keeps the downloaded files and your chats. Use it when the device is short of memory. The model loads again from the saved files.
Remove download deletes the saved model files (in Advanced model controls, or Remove files in Storage & backups). The app asks first and says your conversations will be kept. The model can be downloaded again later, and on-device chat is unavailable until it is.
Clearing all site data in the browser removes conversations, settings and attached documents as well as models. Back up first (see restoring a backup).
Backing up and restoring conversations
Back up: Open Transfer chats and backups (under Import export and transfer, or from Storage & backups with Back up chats). Export saved conversations makes a readable file of your conversations; anyone who has the file can read it. Create protected backup makes a file protected with a password of at least 8 characters, holding chats, memory, projects, presets and, if you choose, documents. A forgotten password cannot be recovered, and the people who run the app cannot open the file either. Each review lists exactly what the file includes. Neither file carries provider settings or API tokens: use Transfer settings for those.
Restore:
Choose Import chats or backup (under Import export and transfer) and pick the file. The app also reads ChatGPT and Claude data exports.
For an encrypted backup, enter the passphrase and choose Unlock backup. “That passphrase did not unlock the backup, or the file was changed” means the passphrase is wrong or the file is damaged; try again exactly as you set it.
Review the preview. Add to my current chats (merge) is selected first and changes nothing already there. Replace all my current chats is an explicit alternative and offers to download a backup of the current chats first.
Choose Import. Tick Also restore memory, presets, projects, preferences to bring those back too.
If you forgot your App lock passphrase, the only way back in is to erase the app's data in this browser and restore a backup.
Dictation, voice conversation or read-aloud is missing
Read-aloud, dictation and voice conversation use the speech services of your browser. The controls appear only where the browser offers them, and no button is shown that cannot work.
The microphone is blocked. Allow it for this site in the browser's site settings (in Safari: the aA menu, then Website Settings, then Microphone) and try again.
The browser never lets sites use speech recognition. Some in-app browsers expose it but block it. The app then says “This browser does not let websites use speech recognition, so dictation and voice conversation are turned off here” and hides the controls for 14 days. Open the app in Safari or Chrome instead (see the known issue).
It needs a connection. The browser's speech service may send audio over the network, so dictation and voice conversation can fail offline. Read-aloud voices marked “online” in the voice list may also need a connection.
Where speech is processed is separate from where the reply is written; see where is voice processed?
Known issues
Each entry names who is affected, what you see, its status, when it was last verified and what you can do. Dates are the last time the project recorded evidence, and an entry says when that was a document rather than a fresh test. Anything not listed has not been reported or verified. Release history is on Updates.
iPhone page restarts during loading or long replies
Affected devices
iPhone 15 Pro on iOS 26.6.2 (the device tested). Other iPhone and iPad models have not been checked.
Browsers
DuckDuckGo 26.6 and Safari automation sessions on that phone.
Models
Qwen3.5-2B on-device model.
What you see
The page is replaced and starts again without an error message while the model loads or a longer reply is being written. Downloaded model files stay saved.
Status
Open, under investigation
Last verified
(project investigation log; many sessions complete normally)
Workaround
Reopen the app: saved conversations and drafts remain and the downloaded files are reused. For dependable use on an iPhone, set Where replies run to a connected provider, or use a computer.
Model downloads fail when the phone is nearly full
Affected devices
iPhone 15 Pro on iOS 26.6.2.
Browsers
Safari automation session.
Models
Qwen3.5-2B download (about 1.20 GB).
What you see
Writing downloaded files fails repeatedly with a generic message while the device has almost no free space. The app keeps retrying and keeps the files it already verified.
Status
Explained by device storage; the app retries automatically
Last verified
(USB system log recorded the out-of-space error)
Workaround
Free space on the phone before downloading. The project's phone tests used at least 3 GB free as the minimum before a download.
Dictation and voice conversation missing in some in-app browsers
Affected devices
iPhone.
Browsers
DuckDuckGo on iPhone exposes speech recognition but does not let websites use it. Other in-app web views may behave the same.
Models
Any (speech is separate from the model).
What you see
A message says the browser does not let websites use speech recognition, and the dictation and voice controls are turned off. They stay hidden for 14 days after the failure is seen.
Status
Browser limitation; the app hides the controls
Last verified
(release v0.11.0 documentation; not re-tested on a physical phone)
Workaround
Open Private Local Chat in Safari or Chrome, which support speech recognition.
Public. Issues on GitHub are public and need a GitHub account. Leave out API keys and private conversations, and review screenshots and error messages before you share them. For setup or performance problems include your device, browser and model.
The source code and release downloads are on GitHub.
