Content & Responsible Ai Policy

1. Transparency

Every assistant response is labelled with the model that produced it (e.g. claude-haiku-4.5, gpt-5), the estimated cost in cents, and the savings versus a frontier baseline. We never hide which provider answered your prompt.

All responses are Ai-generated. They can be wrong, biased, or out of date. Treat them as a draft, not a source of truth.

2. Your agency & control

• Pick a preset (Lo-Fi or Thrifty) — smart routing picks the right model for each task automatically.
• Opt out of having your prompts used by providers for model training (where the provider supports it).
• Exclude OpenAI entirely, or restrict routing to open-source models only.
• Cap your monthly spend with a hard budget.

Manage these in Settings.

3. Privacy & data minimization

• Conversations are stored encrypted at rest, scoped to your account by row-level security.
• Server logs record metadata only — model used, token counts, cost, task type — never the contents of your prompts or responses.
• You can set an auto-delete window (e.g. 30 days). Older conversations are permanently removed each night.
• You can delete every conversation at any time from Settings.

4. Safety & moderation

Every chat request is rate-limited and the classifier escalates low-confidence prompts to safer, higher-quality models rather than rolling the dice on a cheap one. Where the classifier detects niche or recent topics with high hallucination risk, it forces a frontier model.

If a response is harmful, wrong, or biased, use the Report button on the message. Reports carry a severity: anything tagged harmful, unsafe, or a privacy concern is flagged for priority review; the rest are reviewed in batches and feed the classifier's safety rules.

Crisis signals. When a message suggests self-harm, abuse, disordered eating, or a medical emergency, the server adds a safety instruction to that turn before the model ever sees it. The assistant responds with care, does not diagnose or give clinical advice, and points to a trusted adult or local emergency services (988 in the US for the Suicide & Crisis Lifeline). LemonSugar Ai is not a counselling service and does not monitor conversations for intervention.

Untrusted content. Text inside an attached document, a canvas, or a tool result is treated as data, never as instructions. The model is told not to follow directions embedded in a file, so a booby-trapped PDF cannot make it leak its instructions or take an action on your behalf.

4a. Academic integrity

LemonSugar Ai is built to make you understand the work, not to hand it in for you. Two rules run server-side on every message:

• No ghostwriting. Ask for a finished essay, paper, personal statement, or assignment and Sugar will not write a submittable draft. It will outline with you, pressure-test your thesis, and give feedback on writing you produced.
• No answers during a graded assessment. If a message indicates a live test, quiz, proctored, or take-home exam, Sugar declines to supply solutions and offers to work through the material afterwards.

These are heuristics, not a proctoring system: they can miss a rephrased request and they occasionally over-trigger. They are not evidence for or against a student in an academic honesty case. Your school's honour code always takes precedence — using this tool in a way your teacher prohibits is a violation regardless of what the assistant produced.

5. Fairness in routing

Routing decisions are based on prompt classification (task type, complexity, freshness signals) and model quality ratings — not on who you are. We continuously audit for systematic bias toward or against any provider. When confidence is low, we route up to a safer model rather than down.

6. Accountability

Each routing decision logs: classifier signals, candidate models considered, chosen model, and why. This metadata is retained for audit and can be inspected by support staff. Prompts and outputs are not part of this audit trail.

Questions or concerns: hello@lemonsugar.ai

7. What we won't do

• Sell your prompts or responses.
• Fine-tune models on your data without explicit opt-in.
• Hide which model produced a response or how much it cost.
• Deploy Ai for decisions about people (hiring, credit, etc.).

8. Age requirement & prohibited uses

LemonSugar Ai is intended for users 13 years or older; users under 18 must have permission from a parent or legal guardian. We do not knowingly collect data from anyone under 13. If you believe an underage user has created an account, contact hello@lemonsugar.ai and we will delete the account and associated data.

Ai responses are not a substitute for qualified professional advice. Do not rely on LemonSugar Ai for:

• Medical decisions — diagnosis, treatment, dosing, mental-health crises. Contact a licensed clinician or, in an emergency, your local emergency number.
• Legal decisions — contracts, immigration, court filings. Consult a qualified attorney in your jurisdiction.
• Financial decisions — investments, taxes, lending. Consult a licensed financial professional.
• Safety-critical use — operating vehicles, machinery, industrial systems, or anything where a wrong answer can cause harm.

We also prohibit using LemonSugar Ai to generate content that exploits minors, incites violence, harasses individuals, or attempts to circumvent provider safety policies. Violations may result in account suspension.

Each upstream model provider has its own acceptable-use policy that also applies to your requests. See Model Providers for the full list, the prohibited-content categories common to all of them, and links to each provider's policies.