Loading Amethyst…
GPT, Claude, Gemini, Llama, and local models — switch anytime, mid-conversation.
Generate and edit images with Flux and more, right alongside your chats.
Swarm, Debate, and Council — several models working the same problem together.
Build and iterate on real, working apps without leaving the chat.
✦ Demo only — no real payment is processed. Data stays on-device.
All plans run fully on your local hardware. No cloud required.
Plans are saved locally and persist across sessions. Message counts reset every 24 hours.
The agent will always follow your system prompt, plus automatically comply with any additional instructions you give during the chat.
This will permanently remove this conversation from your history.
All models available through this platform are sourced from local or third-party providers.
Requests submitted through the public tunnel's "Request an Account" form. Approving creates the account with the password they chose; rejecting just discards the request — nothing is created either way until you decide.
This will delete your account and all settings and chats. Are you sure you want to do this?
Last updated: August 3, 2026. This policy explains, in full detail, how Amethyst collects, uses, stores, shares, retains, and protects information in normal use, private/temporary chats, review situations, and account deletion — and what choices and rights you have over your own data. New accounts must agree to this policy to sign up; it does not apply retroactively to accounts created before this date.
This Privacy Policy applies to your use of Amethyst (the "app"), including its chat interface, account system, Image Studio, Code Agent, Arena, Crystals, voice and live-conversation features, support chat, and any other feature offered within it. It covers information collected through the app itself, and information the app sends to third-party AI model and service providers on your behalf when you choose to use them.
It does not cover the privacy practices of third-party websites, apps, or services you may reach through links in the app, or of the AI providers themselves beyond describing what is sent to them — each provider has its own privacy policy governing what it does with data once it receives it, and we encourage you to review those policies for the providers you choose to use.
By creating an account, requesting an account, or continuing to use this app, you agree to the practices described in this policy. If you do not agree, please do not create an account or submit personal information to the app.
The sections below go into full detail, but here's the gist up front: we don't sell your data, we don't run ads, and we don't use advertising trackers of any kind. Passwords are never stored in plain text — only a one-way hashed record that can't be reversed, even by us. You can delete individual chats, clear saved memory, or delete your entire account at any time from directly within the app, no waiting on a request form.
The honest tradeoff: this app is built and run by a single independent operator, not a company with a dedicated privacy or compliance team. That means some things a large platform would automate — a self-service data-export button, a support ticket queue with guaranteed response times — don't exist yet. What you get in exchange is a small, direct line to the person actually responsible for your data, and a policy that describes what really happens rather than what a legal template says should happen.
Amethyst is an independently operated application, run by a single operator rather than a corporation. There is no separate privacy office, legal department, or support team — the operator who runs the service is also the one who reads any message you send through Support & Help. Where this policy refers to "we" storing, reviewing, or protecting information, it means the operator and the software they run.
This is a function of scale, not corner-cutting: the same architecture that keeps this app small and direct also means some formal mechanisms found at large platforms don't apply here. Where this policy describes a right or a choice you have, the practical way to exercise it is through the app's own settings and account controls, or by reaching the operator directly (see Section 21).
We collect information in a few different ways, described below.
Information you provide directly. When you create or sign in to an account, the server stores account data so your experience can be restored later. This may include your username, a password record (never the plaintext password itself — see Section 17), an email address if one is provided for account requests or verification, settings, plan and usage counters, saved chats, user instructions, memory-bank facts you've asked the app to remember, generated image references, and other saved app state.
Content you submit for processing. This includes chat messages, pasted document bubbles, uploaded images, voice recordings and live-conversation audio, code and files submitted to the Code Agent, and any prompts, instructions, or context you provide so the app can generate a response.
Technical and device information. The server may store an IP association file for account/session convenience, so the same device can resume the correct account without signing in again. Standard web request metadata (such as request timestamps, endpoint paths, and response status) may also be recorded as part of normal server operation and logging, described further in Section 14.
Information generated through use. This includes Arena battle votes and resulting Elo-style rankings, image generation history shown under "My Stuff," credit/usage counters, and similar records of how you've used the app's features.
We do not intentionally collect sensitive categories of information (such as government ID numbers, financial account details, or health records) as part of normal account creation, and you should avoid submitting this kind of information in chats or file uploads — see Section 6 for why.
We use the information described above to: create and maintain your account; authenticate you when you sign in; save and restore your chats, settings, and preferences across sessions; route your messages and content to the model or provider you select so it can generate a response; enforce usage limits and rate limits so the app remains available and is not abused; diagnose bugs, failures, and server-health issues; protect account saves from corruption during writes; investigate suspected abuse, fraud, or security incidents; and comply with legal obligations described in Section 12.
We do not sell your personal information, and we do not use your chat content to run advertising.
You are in control of the account-level decisions that matter most, directly from within the app itself — no request form, no waiting period:
Because this app is operated by one person rather than a company with a formal rights-request pipeline, exercising anything beyond the in-app controls means messaging the operator directly through Support & Help — which, in practice, tends to be faster than a corporate ticket queue.
Your messages, pasted document bubbles, images, voice/live-chat content, and other prompt context are sent to the cloud AI provider running the model you select, so the app can generate a response.
Depending on which features and models you choose to use, this may include cloud AI providers such as Mistral, Google Gemini, Cohere, NVIDIA, and OpenRouter (which itself may route to a variety of underlying model and speech providers, including Fish Audio for certain voice options); web search via Tavily and Google's favicon service (used only to fetch site icons); image generation via Pollinations; and GitHub, when a feature you use integrates with it. Each of these is a separate company with its own infrastructure and its own privacy practices; once your content reaches one of them to generate a response, it is subject to that provider's own data handling, consistent with the terms under which the operator uses their service.
Web search, image generation, music generation, speech, and support-chat features may send the specific information needed for those features (such as a search query, an image prompt, or text to synthesize into speech) to their configured services. We select which providers are available and configure how requests are routed to them; we do not control what each provider does internally with a request after it is sent.
The app uses browser cookies and/or session storage to keep you signed in between visits, remember interface preferences (such as theme and layout choices), and maintain security tokens (such as CSRF protection tokens) used to verify that requests actually came from your own session. These are functional technologies needed for the app to work, not third-party advertising or cross-site tracking cookies.
We do not use third-party advertising trackers, marketing pixels, or cross-site analytics cookies. Clearing your browser's cookies and site data for this app will sign you out and reset remembered preferences.
Regular chats are saved to your account so they can appear in the sidebar and be restored later. Account saves include the conversation content and the settings needed to recreate the session.
Support bot sessions are designed to run in browser memory for that support session rather than being saved as regular account chat history.
Temporary Chat is intended to live only in memory. It is removed when you exit temporary mode, sign out, close the page, or start a regular chat. Temporary chats are not saved to your account by the normal autosave flow.
Important: Temporary Chat does not make external providers forget requests already sent to them. Once content has left this app for a cloud model or service, exiting Temporary Chat does not recall or erase it from that provider's own systems.
We retain account data, saved chats, settings, and usage records for as long as your account exists, so that the app can continue to function the way you expect it to. Temporary Chats are, by design, not retained beyond the session described in Section 10.
Diagnostic and operational logs (Section 14) are retained for a limited operational window sufficient to diagnose recent issues, after which they may be rotated out or overwritten as part of normal server maintenance; we do not commit to a fixed retention schedule for these logs, since retention needs can change as the app evolves.
Deleting your account, described in Section 13, removes your primary account save going forward, but does not necessarily remove every trace immediately in every location described in that section (for example, existing backups or diagnostic logs already written before deletion).
Your chats, account data, and other information stored by this app can be looked at in legal cases or emergencies. This includes: responding to a valid legal process (such as a subpoena, court order, or law enforcement request); a credible safety threat to you or someone else; suspected abuse, fraud, or an attempt to harm systems or people; and emergency troubleshooting needed to keep the app or your data safe.
Access in these situations is limited to what's reasonably necessary to resolve the matter, and the goal is always to protect users, systems, and other people. This is not a promise that your data is otherwise reviewed at will, and it does not mean routine, everyday use of the app is monitored.
Using Delete account removes your account save file and its password sidecar record. If the account folder becomes empty, the server attempts to remove that folder as well.
Deletion may not remove separate diagnostic logs, generated media files, provider-side records, browser downloads, exports, backups, or data that already left the app before deletion. If you would like help confirming what remains after deletion, you may reach out through the app's Support & Help feature.
The app and server may write operational logs such as sign-in, save, delete, support, image-generation, speech, or server-health messages. These logs help diagnose failures, protect account saves from corruption, and understand whether the app is working correctly. They are used for operational purposes only, not for profiling or advertising.
This app is not directed at children, and we do not knowingly collect personal information from children under the age of 13 (or the equivalent minimum age in your jurisdiction). If you believe a child has created an account or submitted personal information to this app, please contact the operator through the Support & Help feature so the account and associated data can be reviewed and removed.
Account data, saved chats, and logs are stored on the operator's own server infrastructure and are not routed through third-party cloud storage as part of normal account operation. When you use a cloud AI model, search, image-generation, or speech provider (Section 7), your relevant content is sent to that provider's own infrastructure, which may be located in a different country than you or the operator, and processed under that provider's own data-handling and international-transfer practices.
Account passwords are not stored in plaintext. The app stores a password record produced by a one-way hashing process, so that the original password cannot be recovered from the stored record itself. Even the operator cannot simply "look up" your password from account storage.
Some browsers offer a "Do Not Track" signal. Because this app does not use third-party advertising trackers or cross-site analytics in the first place (Section 8), there is no separate advertising-tracking behavior for such a signal to turn off, and the app does not currently respond differently to it.
This policy may be updated as the app's features, providers, hosting setup, or legal obligations change. When it is updated, the "Last updated" date at the top of this document will change. Material changes that affect how new accounts must agree to this policy will be reflected in the version and acceptance timestamp stored with newly created accounts; this policy does not apply retroactively to accounts created before a given update unless you are separately notified and asked to re-agree.
We actively and regularly work to find and close security gaps and vulnerabilities in this app, and treat a successful attack against it as unlikely. Still, honesty requires saying plainly: no system connected to a network — including this one — can be made perfectly, absolutely secure, and that's true no matter how much care goes into it.
Because of that, by creating an account, you're acknowledging the same thing anyone using an internet-connected service implicitly accepts: that no guarantee of absolute security exists, and that the operator of this app isn't liable for a cyberattack, unauthorized access, data breach, data loss, or other security incident affecting the app or your data — whether the cause is a flaw in this software, a third-party provider it connects to, the network or hosting environment it runs on, or anything else outside the operator's direct control.
This doesn't remove protections that legally cannot be waived under applicable law; it describes the operator's position on liability to the fullest extent the law allows, and it doesn't change how seriously security is taken day to day — see the first paragraph above.
This policy describes the behavior of this app as configured here. If the server, providers, tunnel, hosting setup, or code changes, privacy behavior may change too. This document is meant to be readable product disclosure, not legal advice, and it is not a substitute for consulting a lawyer about your own specific situation.
If you have questions about this policy, or about the information associated with your account, the way to reach the operator is through the app's Support & Help feature.
Agent Aura, Agent Prism, Agent Cipher, Agent Zypher, Agent Photon, Agent Crouton, and Agent Crumb do not support image input. Switch to Agent Sonata, Agent Lyric, or Agent Swift to upload images.
Good to see you again.
Join Amethyst in a few seconds.
Chat, create, and code with the best cloud and local models — all in one place.
Your session ended, but this chat is still here — sign back in to keep saving it.
Lives only in memory — never saved to any server. Use the Export button to save it before you leave. Start a regular chat or close the page and it's gone, leaving no trace.
Copy the prompt below into a chat with your other AI assistant, then paste its reply back in here.
Model Tester status means the Arena tab is free for you — every vote you cast deposits a credit into your account instead of spending one. No catch, just testing.
Please put it to use: head to Arena and start battling models. Bonus — the Leaderboard tracks a Contributors ranking too, of accounts with the most votes cast. Ironic, given that's literally the leaderboard tab: the more you test, the higher you climb on it.
Beta Tester Mode is for people helping evaluate Arena inside Amethyst. It is granted per account after review, so submitting the form does not enable it instantly.
Arena stops consuming your normal credits. Each Arena round reverses the flat Arena cost instead, so beta testing does not eat into your regular account budget.
Your account gets the Beta Tester badge in Settings and Arena, a one-time welcome modal the first time the status is granted, and a tester chip beside your name in the Arena Contributors leaderboard.
Spend time in Arena comparing models and casting votes. The Contributors view tracks lifetime Arena votes, so more testing activity shows up there over time.
All messages and responses in this session are being recorded and monitored. By continuing you acknowledge this.
Please contact your local administrator about using Music's LLM.
Cloud models ranked by Elo rating from your Arena votes. Models flagged "Frequently Unavailable" are excluded from battles so routing, rate limits, and queueing do not distort model comparisons.