Cassia starts with a deterministic core. Stockfish, a chess engine running in your browser as a WebAssembly worker, evaluates each position. A rule-based pipeline turns those evaluations into insights (for example: hung piece, missed tactic, king exposure) and renders them with templated text. The facts you rely on — the evaluation, the best move, the classification — are computed by the engine and the rule-based pipeline, never by a language model.

Engine analysis runs locally in your browser. The positions you analyse are not sent to our servers for that analysis. For a small number of position lookups the app may query the public Lichess Cloud Evaluation API to check for an existing community-shared evaluation.

An optional layer (internally called the “Rephraser”, and any AI-assisted summary) may use a third-party large language model to rephrase or expand the deterministic coach text. When it runs, the model does not pick the chess moves or compute the evaluation; the deterministic engine does. The model's job is to rephrase facts that are already established, within constraints designed to prevent it from inventing analysis.

When the AI-assisted layer runs, the request may include: the position (FEN), recent move history, engine evaluation context, the deterministic Cassia text that would otherwise be shown, and limited technical metadata needed for rate limiting and abuse prevention. We avoid sending information that is not needed to produce the response, and we do not send your name, email, or account identity to the AI provider.

We reach these models through a third-party AI routing provider (currently OpenRouter), which forwards the request to an underlying model provider. Which providers and models are used — and their data terms — depend on your plan and on our current routing configuration, and may change over time. We favour routing to providers that do not use your inputs to train their models and that retain request data only for a limited period (for abuse prevention and reliability) rather than indefinitely. Content moderation and acceptable-use enforcement for AI output are our responsibility, not the underlying provider's. Because the specific provider can vary, this describes the policy we hold providers to rather than the terms of any single vendor.

The set of approved AI providers is controlled by zen.reinit.net and may change. When it changes, this page and the Privacy Policy are updated.

When you flag a Cassia response or submit feedback, we store the position, the relevant moves, the classification, the full coach text that was shown to you, any reason you type in, and your account identifier. We use this feedback to investigate failures, fix the deterministic pipeline, and (when AI-assisted output is enabled) tune prompts and provider routing. We may also log request metadata (timestamps, latency, provider outcomes, error categories) for abuse prevention, quota enforcement, and debugging.

Do not include sensitive personal data, secrets, or third-party private information in feedback text, annotations, or any future AI prompt input.

Cassia output, engine evaluations, classifications, and any AI-assisted rephrasing are study aids. They can be incomplete, stale, or wrong. They are not authoritative chess analysis, tournament advice, or professional coaching, and must not be treated as such. You are responsible for independently verifying anything you rely on, including by stepping through the engine's suggested lines.

AI-generated text in particular can hallucinate variations, misread positions, or describe moves that do not exist. The deterministic pipeline is designed to constrain what the model can say, but no constraint is perfect. If you see something wrong, please use the feedback button so we can fix it.

By using the Service you consent to the processing described above. If you do not want your input processed by a third-party AI provider, do not use AI-assisted features when they become available; the deterministic deterministic explanation layer will remain accessible.

When the production behaviour of Cassia changes — for example, when an AI provider is first enabled, when a new provider is added, or when logging practices change — this page and the Privacy Policy are updated and the “Last updated” date above is revised.

Questions or concerns about how Cassia works can be sent to [email protected].