Privacy
The four things tryingopen keeps about your visit
A plain reading of the tryingopen privacy policy and terms, covering the few things Offline Base keeps and what stays in your browser or on your device.
Offline Base, the team behind tryingopen, keeps four things when you use the tryingopen chat: an hourly message count per IP address, page view counts, hosting logs, and the size and timing of requests on our edge devices. Everything else you set up or make stays in your browser, on your device, or in storage you connect.
That list comes from the tryingopen privacy policy, effective October 12, 2026.
No accounts, no cookies, no message logs
For the chat, the policy's opening list and the sections after it come down to this:
- The chat has no accounts, so you never give a name or an email to use it.
- "The chat sets no cookies."
- "We do not log the text of your messages or of the replies."
- "We do not store your conversations on our servers."
- "We do not sell your information, use it for advertising, or use it to train AI models."
A message goes from your browser to our server, which passes it on to be answered and streams the reply back without writing the text down.
What the company serving the model keeps depends on the model and provider you pick. What zero data retention covers, and where it stops maps those routes.
What we keep
The "What we keep" section lists three things. A fourth sits in the paragraph about our own hardware.
| What | Held by | How long | What it contains |
|---|---|---|---|
| Messages sent per IP address, per hour | Upstash | Expires on its own within about two hours | A count for each IP address |
| Page views | Vercel Web Analytics | Not stated | Visits, counted without cookies and without identifying you |
| Hosting logs | Vercel | "A limited period" | Things like IP address, time and page requested; no messages |
| Request records on our edge devices | Offline Base | Not stated | Size and timing of each request; no text |
A count per IP address
To keep the free service open to everyone, we count each IP address's messages per hour. Upstash holds those counts, and they "expire on their own within about two hours."
Page views
Vercel Web Analytics counts our visits "without cookies and without identifying you."
Hosting logs
Vercel hosts the site and keeps standard request logs (IP address, time, page requested) "for a limited period." Those logs "do not contain your messages."
Size and timing on our edge devices
The default pick on tryingopen.com is labelled "Runs on an edge device". NVIDIA Jetson Orin Nano boards that Offline Base runs answer the quick questions sent to it. The policy says the boards record each request's size and timing. The reply's "Learn more" panel adds that the fleet logs each request's model, token counts and timing.
If you pick a cloud model, your first two questions also go to our devices for the side-by-side, with the same size and timing record. How the side-by-side compares our small model with your pick shows what that panel does.
The boards are reached through an ngrok tunnel, as the policy notes. The router, d1, a small decision model from Liquid AI, picks between the board and the cloud. It runs on one of our own devices, so no other company receives your message to make that choice. How the d1 router picks between our small model and the cloud covers how it decides.
That is the full list of what Offline Base holds.
A separate section of the policy covers the optional developer API at tryingopen.com/developers. It lists what an account would keep: an email and a password hash, a 30-day sign-in cookie, the names you claim, and usage records (time, model, provider, status, tokens and credits) for each name's last 20 requests for up to 90 days. Upstash holds it, and none of it includes messages or replies. As of October 14, 2026, accounts aren't open yet.
What stays with you
Settings and presets
Your browser keeps your theme, your picks in the composer and any notice you closed in its local storage. Presets work the same way, as the preset menu says: "Presets are model setups saved in this browser. Each one has a link that opens it." To give someone a preset, you send its link.
Share links
A share link carries the conversation after the "#", a part of the address browsers never send to any server. We hold no copy, and anyone with the link can read it. The model setup sits in the ordinary part of the address, which is sent like any page request. Sharing a chat, a model setup or a preset as a link has the details.
Self Chat's local model settings
In Self Chat you can connect a model running on your own computer in Ollama, LM Studio, llama.cpp or Jan. Your browser sends messages straight to it, and they never reach our servers or the internet, unless you switch on web search. Then your latest message goes to our server for one lookup, and only the results go back to your computer. To bring the model back next time, "your browser remembers the server's local address and its model names." Use Ollama, LM Studio, llama.cpp or Jan from Self Chat walks through the setup.
Self Chat's conversations go to storage you connect, which as of October 14, 2026 means Google Drive or a folder on your computer. The policy also covers OneDrive and Dropbox "where offered"; the Self Chat page lists them as coming soon. A chat saved to Google Drive is a file Google holds under its own privacy policy. Self Chat saves your chats only where you choose covers that part.
Files, previews and exports
Some models in the picker can create files such as web pages, documents and CSV tables, each previewed beside the chat. The Export button saves the conversation as tryingopen-chat.md. Files the model creates, their previews and exports "are put together in your browser and saved straight to your device." Sending pictures and PDFs, and getting files back has more.
To remove everything the site keeps in your browser, clear this site's data in your browser settings.
What the Terms of Service add
The Terms of Service are effective October 3, 2026. Their section on the service opens with: "Tryingopen lets you chat with open AI models for free, with no account."
Two parts matter most from day to day.
AI answers can be wrong. The terms say replies "can be inaccurate, incomplete, out of date or offensive." Check anything important, and take medical, legal or financial questions to a professional.
The service can change. Offline Base "may change, limit or stop any part of the service at any time, including which models are offered and how many messages you can send." What's new in the picker since September tracks recent changes.
When the policy changes
The privacy page promises: "If we change what happens to your information, we will update this page and the date at the top." The current date, October 12, 2026, came when the default pick's routing moved onto our own hardware and the developer API section was added. The date at the top of tryingopen.com/privacy tells you which version applies.