Chat privacy
This page describes how the current KAMUI Chat application handles sign-in and conversations.
Google sign-in
When you sign in with Google, KAMUI Chat receives your name, email address, and Google account identifier. These details are used to identify your session, display your account, and determine administrator access. Google handles your Google password; KAMUI Chat does not receive it.
Cookies and sessions
KAMUI Chat uses essential cookies for sign-in and session security. Session records contain your account details and expire after eight hours. Expired session and sign-in transaction records are deleted during cleanup on a subsequent accepted sign-in attempt. Temporary sign-in security cookies last ten minutes. A successful sign-out deletes the current stored session and clears its session cookie.
Chat messages and model providers
Your prompt and the conversation context included with it are sent to the third-party model provider serving your selected KAMUI model to generate a reply. KAMUI Chat does not automatically add your Google name, email address, account identifier, or sign-in credentials to model requests. Information you type into a message is still part of the submitted prompt.
The current Chat application keeps conversation history in browser memory and does not provide saved conversation history. Reloading the page or starting a new chat clears the conversation held by the page. Private model connections may create temporary execution records while a request is processed; hosting services and model providers may retain records under their respective policies.
Scientific Simulation
For the existing hosted scientific tools, your messages are processed by the selected language model as part of Chat. After the model has gathered the numerical conditions, KAMUI forwards only the validated experiment parameters to its hosted physics solver. Google account details, sign-in cookies, credentials, and the conversation text are not forwarded to the physics solver. A compact result summary, including the parameters, run identifier, and reported numerical measurements, is included in subsequent Chat context so the model can discuss the result. Returned fields and conversation context remain in browser memory and are cleared when you reload or start a new chat.
For an explicit on-device PDE learning request, Chat prepares the equation, Cartesian domain, coefficients and boundary/initial conditions in your browser. The original Poisson example remains available; the expression-based experiment accepts derivatives through order two with at most three coordinates and two fields. Missing conditions must be supplied before running. Choosing Run on this device confirms the displayed conditions and starts the small network in a native browser Worker. Advanced settings are optional. This action makes no model or training API call; the local adapter does not post network weights or predicted fields to a server. Conditions and compact measured text summaries remain in this conversation. If you later send an ordinary Chat message, those text conditions and summaries may be included in the conversation sent to your selected model provider. The existing hosted scientific tools described above remain separate cloud computations.
Financial risk scenarios
For an explicit BTC/ETH account-risk request, KAMUI sends the public account address you supply to Hyperliquid to read its current positions, market data and fee tier. You can enter an address or select it from your browser wallet. This does not verify account ownership or request a signature, investment or trade. KAMUI calculates the agreed scenario on its server. After the snapshot arrives, interactive risk charts explore linear price shocks using the same observed positions and cost estimates. Chart adjustments stay in your browser and do not place orders or refresh market data. The address and raw account response are excluded from subsequent language-model context; a compact calculation summary may be included so the model can discuss the result. Scenario cards and conversation context remain in browser memory until the page is reloaded or the chat is reset. Hyperliquid and hosting providers handle requests under their own policies.
Usage and service operation
KAMUI Chat stores sign-in, Chat, and simulation usage counters to apply service limits. These counters use a hashed connection identifier derived from the request IP address. For kamui1-gpt, a separate hashed Google account identifier also enforces hourly and daily account limits; the account identifier and your email address are not sent to its model provider. The current application has no scheduled deletion of Chat or simulation usage counters. Administrator access is determined on the server from the authenticated account.
Persistent neural companion
KAMUI runs a small connected spiking-neuron simulation in a dedicated Worker on your device. Actual modeled sensory responses drive an explicitly engineered inertial screen body; body pose and visual cues affect subsequent input. Signed source connections are preserved. Neuron dynamics, input transduction and body mechanics are declared modeling choices, not a validated biological fly, aerodynamic body or language decoder. The neural clock advances through calculated steps at a measured rate; no recorded flight loop is substituted. Browser-origin storage retains its identity, neural state, delayed spikes, random-generator state and body state across normal reloads. A single browser lock coordinates visible tabs. Hidden pages hibernate; closing the site does not continue an unseen cloud brain. Normal checkpoint recovery can lose the interval since its last successful save; browser storage clearing removes the saved organism. No provider credential or account identifier is copied into the Worker. A bounded raster of only the message text becomes a numerical visual profile for the modeled visual input cells; it is an engineered mapping, not biological reading or semantic understanding. The matching completed neural measurement and original question are sent to the selected language service to generate words. The local checkpoint may temporarily contain numeric visual-input features and their checksums, but never a raw chat transcript or provider credential. Client-computed measurements are not server attestations. Essential same-origin asset cookies may be attached automatically by the browser. Existing explicitly requested hosted scientific experiments keep their own execution limits and costs.
Account activity measurement
From the measurement start date, KAMUI records successful Google session issuance, daily authenticated use, and a five-minute activity lease for signed-in accounts. These records support aggregate product and investor reporting. The analytics record uses a dedicated HMAC pseudonym derived on the server from the verified Google account identifier; it does not copy your email address, name, Google identifier, session cookie, prompt, wallet address, or IP address into the engagement tables. Administrator activity is classified separately from external-user metrics.
The activity lease expires after five minutes without a visible-page heartbeat. Detailed login and daily activity rows follow a 400-day retention policy with scheduled cleanup; the pseudonymous account roster is retained for cumulative unique-account reporting. These measurements do not count anonymous visitors and do not prove that a person was continuously attentive. Contact us to request deletion of account-linked analytics data.
Your choices and contact
You can sign out of Chat or revoke KAMUI's Google connection in your Google Account settings. Revoking the Google connection does not immediately end an existing Chat session. Sign out of Chat as well; otherwise that session remains valid until its eight-hour expiry.
For questions about Chat privacy or your account data, contact kamuitranslator@gmail.com.