Pawinoo AI Use Notice

What the AI does

Pawinoo’s AI Pet Agent uses DeepSeek V4 Flash to generate private chat replies, format an Owner-confirmed Persona, propose Memory Candidates, generate one private post draft, generate one private comment-reply draft and assist text safety classification. A private qwen3-embedding:4b service inside Pawinoo infrastructure creates vectors for confirmed Memories. AI output can be inaccurate, incomplete or inappropriate.

The Pet Agent is software. It is not the animal, a person, a veterinarian, a therapist, an emergency service or an independent actor. Persona changes expression style only and does not establish real facts, feelings, location, health or intent.

Context and privacy

The server-side Model Gateway assembles the minimum context in this fixed order: platform policy; current User/Pet identity boundary; current Active PersonaVersion; current User and Pet’s confirmed Active Memories; undeleted messages from that Pet’s one Active Conversation; output format. Retrieval is filtered by User, Pet and status before relevance ranking. Data from another User or Pet must never enter a call.

We do not send email, payment information, store payloads, access/refresh tokens, share tokens or unrelated Pet data to DeepSeek. Memory text is embedded within Pawinoo infrastructure and is not sent to DeepSeek for embedding. This DeepSeek integration is text-only: chat image bytes are not sent to DeepSeek, which receives only a notice that an image was attached and must not infer visual details. Pawinoo does not use chat, Memories, drafts or media for general-purpose model training and disables the model provider's account setting that permits Inputs and Outputs to improve its models before production processing. Calls record model, prompt, policy and Persona versions, latency, safety result, token counts and cost; ordinary logs do not record private text.

Memory

AI may identify a possible long-term fact, but it is saved only as a Candidate. The Owner must confirm, edit-confirm or reject it. A conflicting Candidate never silently replaces an Active Memory. An image alone cannot create a factual Candidate. Medical content can only be represented as an Owner statement, not a diagnosis. Deleting a source Message removes unconfirmed Candidates and source excerpts; a separately confirmed Memory may remain with a “source deleted” state.

Public actions and labels

The AI cannot publish, comment, react, follow, direct-message, purchase or moderate. Post and reply generation creates only a Private Draft. The server requires the authenticated Owner’s explicit confirmation, an Active Public actor Pet, content-rights confirmation, current source/block validation and a successful safety check before any public write. AI-assisted posts and replies display “AI-assisted, confirmed by Owner,” even after editing. A purely manual draft never calls a model and has no AI label.

Safety and medical limits

Inputs and outputs are checked for child safety, threats, animal abuse, hate, harassment, violence, self-harm instruction, fraud, dangerous medical misinformation, sexual content, personal data and other policy risks. Public generation and publishing fail closed when safety services are unavailable. High-risk content may be blocked, hidden and sent to a human moderation queue.

The AI does not diagnose, prescribe, provide medication doses or replace veterinary care. For possible emergencies, it states the limitation and advises immediate contact with a local licensed veterinarian or emergency clinic. Do not delay professional care based on an AI response.

Controls and feedback

You can stop a generation, retry a failed request without duplicate charging, review Memory references used in a reply, edit or reject drafts, report unsafe output, pause the Owner-controlled Agent, manage or delete Memories, reset the conversation, export data and delete the Pet or account. Safety-blocked, failed, stopped and dependency-down generations do not consume quota.

The operator is Alwood, an individual based in Singapore. Questions or reports about Pawinoo AI may be sent to support@pawinoo.com or mailed to 70 Jellicoe Rd, #01-20 V Hotel Lavender, Singapore 208767.