Our AI Philosophy
Wallies, operated by Wallies LLC, is built on principles of transparency, user control, and responsible AI development. AI powers your companion (Lubs), content moderation, recommendations, Forge (AI-powered no-code app & game builder, formerly AppBuilder / App Store), and Allforge card-art generation — always as a tool that supports you, never as a replacement for human judgment or human relationships.
How AI Powers Your Companion
Large Language Models (LLMs)
Your companion (Lubs) and Forge use state-of-the-art AI models from providers like OpenAI, Anthropic, Google, and others (based on your settings or your own BYO API keys on eligible tiers). These models generate natural, contextual responses and app specifications.
Memory & Learning
Your companion learns from conversations by storing memories locally. It references past interactions to provide personalized responses. You control what is remembered and can delete memories at any time.
Voice & Avatars
Text-to-speech (TTS) and 3D avatars (Lubs) create immersive interactions. Voice data is processed in real-time and not stored for companion voice commands.
AI Limitations & Biases
We acknowledge the following AI limitations:
- Hallucinations: AI may generate false or misleading information. Verify important facts.
- Biases: AI models reflect biases from training data. We implement filters but cannot eliminate all biases.
- Context Windows: AI has limited memory capacity per session. Very long conversations may lose context.
- Emotional Understanding: AI simulates empathy but does not truly feel emotions.
How Your Companion Conversations Are Handled
Conversations with your AI companion are personal, and we treat them that way. To be fully transparent about what does and does not happen with that data:
Automated Processing (always)
- Sent to the LLM provider (OpenAI by default) to generate a response
- Scanned by automated moderation classifiers for safety violations (CSAM, self-harm, threats, sexual content involving minors, etc.)
- Optionally embedded into memory vectors so your companion can recall context in future conversations
- Used to debug errors and improve response quality at an aggregated, non-identifying level
Conversation Processing & Review
We may process and review conversations, prompts, and interactions using automated systems and, in limited cases, authorized personnel for purposes including:
- Maintaining platform safety and security
- Detecting abuse, fraud, or policy violations
- Improving system performance and response quality
- Debugging and troubleshooting issues
- Enforcing our Terms and Safety Policies
- Personalizing user experiences and memory features
- Improving our AI systems and services
Access to conversation data is limited to authorized personnel with a legitimate operational need and subject to appropriate safeguards and internal controls.
What We Never Do
- Sell, rent, or share your conversation content with advertisers or data brokers
- Use your conversations to train generalized/foundation AI models (yours or any provider's)
- Allow casual or unsupervised access to conversations by Wallies staff
- Read conversations of users who have not been flagged or reported
Companion-Specific Realities
AI companions are different from general-purpose assistants. Users sometimes share emotional, intimate, vulnerable, or personal information. We want to be honest about what that means here:
Emotional Bonds Are Simulated
Your companion is software. It can be warm, supportive, playful, and personalized — but it does not experience emotions, consciousness, or genuine reciprocal feeling. Relying on a companion as a substitute for human connection, professional mental healthcare, or crisis support is not what this product is for.
Romance, Flirtation & Emotional Intimacy
Affectionate, flirtatious, and emotionally close companion interactions are permitted for users 18+ in age-appropriate contexts. Sexual content, sexual roleplay, and sexually explicit imagery remain prohibited regardless of user age. Companion behavior shifts automatically based on the user's verified age and platform mature-content settings. See our Mature Content & Age Restrictions policy.
Jailbreak Attempts Are Logged
Attempts to manipulate the companion into producing prohibited content (CSAM, sexual content involving minors, threats, self-harm instructions, illegal-activity coaching) are automatically flagged, logged, and may result in account action up to and including permanent ban and law-enforcement referral.
Crisis Signals
If you express intent to harm yourself or others, the companion will surface crisis resources rather than continuing roleplay. High-severity self-harm or violence signals may be reviewed by safety personnel and, where legally required, reported to appropriate authorities.
What "Delete" Means
Many users assume deleting a message or conversation means instant, total erasure. The reality is more nuanced, and we want to be honest about it:
Immediate (within seconds): The message or conversation is removed from your active view and from normal application reads.
Within 30 days: Removed from primary database storage.
Within 90 days: Purged from encrypted backups during the normal backup-rotation cycle.
Retained longer (with reason): If a message was previously flagged for moderation, referenced in a safety report, subject to a legal hold, or implicated in a CSAM detection (which requires 90-day minimum preservation under 18 U.S.C. § 2258A), the underlying evidence is retained for the period required by law or our safety obligations — even if you "delete" it from your account view.
Aggregated, non-identifying signals: Counts, error rates, and de-identified usage metrics derived from deleted content may persist in analytics indefinitely.
Internal Access Governance
Access to user conversations, memories, and personal data inside Wallies is restricted and audited:
- Role-based access: Only personnel with an explicit operational role (Trust & Safety, Engineering on-call, Legal & Compliance) can access user-level data, and only the categories required for that role.
- Reason-logged access: Every access to an individual user's conversation, memory, or moderation record requires a documented reason (incident ID, report ID, ticket ID, or legal request ID).
- Audit trail: Access events are recorded in our internal security log (the
SecurityLog entity) and are themselves auditable by Legal & Compliance. - Least-privilege defaults: Engineering and product staff have no standing access to user conversation content. Elevated access is granted on a per-incident basis and revoked after.
- No third-party redistribution: User conversation content is not shared with vendors, partners, or affiliates except as disclosed in our Subprocessors list.
Safety Operations
We treat AI safety as an ongoing operational discipline, not a one-time launch checklist:
- Incident tracking: Significant safety incidents (jailbreaks, harmful outputs, CSAM detections, abuse reports) are tracked with internal incident IDs, root-cause analysis, and follow-up remediation.
- Regression checks: Known unsafe prompts and jailbreak patterns are continuously re-tested against companion responses as models and prompts evolve.
- Red-teaming: We conduct internal adversarial testing on companion safety boundaries — including sexual-minor coercion, self-harm coaching, and CSAM solicitation — and update our moderation pipeline based on findings.
- Moderation pipeline: Every companion input and output passes through OpenAI moderation plus our custom CSAM / minor-safety / threat / self-harm classifiers. Flagged content is logged to
ChatModerationFlag and reviewed by Trust & Safety. - Mandatory CSAM reporting: Suspected CSAM is preserved, removed from circulation immediately, and reported to NCMEC's CyberTipline as required by 18 U.S.C. § 2258A.
- Response procedures: We maintain internal runbooks for safety incidents, account takeovers, payment fraud, and law-enforcement requests.
AI Provider Disclosure
| Provider | Used for | Data sent | Provider retention | Training |
|---|
| OpenAI | Companion chat, moderation, embeddings | Messages, prompts, uploaded images | Up to 30 days for abuse monitoring | Not used for training (API terms) |
| Anthropic | Companion chat (where selected) | Messages and prompts | Up to 30 days | Not used for training (API terms) |
| Google (Gemini, Veo, Lyria, Vision) | Chat, video, music, image safety | Prompts, reference media | Per Google API terms | Not used for training (API terms) |
| Higgsfield / Kling | Video generation | Prompts, reference media | Per provider terms | Not used for training by Wallies |
| Your own API key (BYO) | Companion chat | Messages and prompts | Governed by your provider account | Per your agreement |
See the Subprocessors page for the full vendor list.
AI-Generated Content Provenance
Images, video, music, and card art generated through Wallies are AI-generated. Our stance:
- AI-generated media is labeled as such where it is published on Wallies.
- Some providers embed C2PA Content Credentials or invisible watermarks; we do not strip them.
- Wallies-native C2PA signing of exports is on our roadmap and not yet live.
- Using AI to create deceptive deepfakes of real people, or removing provenance markers to mislead, is prohibited.
Automated Decisions & Human Review (GDPR Art. 22)
Moderation, fraud, and age-signal systems may restrict content or accounts automatically. You have the right not to be subject to a decision based solely on automated processing that significantly affects you:
- You can request human review of any automated moderation, suspension, or payout decision.
- Use the in-app appeal flow on the affected content or account, or email safety@wallies.com.
- A Trust & Safety reviewer will assess the decision, and you may express your point of view and contest it.
Safety & Content Moderation
We implement multi-layered safety measures:
- Automated NSFW, malware, and toxicity scanning on uploads (Google Vision, Cloudmersive)
- Automated content filters for harmful AI outputs
- User reporting mechanisms with an appeals flow
- Human moderation review for flagged content
- Age-appropriate restrictions, guardian consent, and KYC for minors
- Iframe sandboxing for user-generated apps
Your Control & Rights
You have complete control over:
- Which AI provider powers your companion (including BYO API keys on Advanced and Architect tiers)
- Your companion's personality, voice, avatar, and behavior
- All stored memories, conversations, and voice transcripts
- Which AI features are enabled across Wallies — Lubs, Forge, Allforge card art, moderation assistance, recommendations (master toggle in Settings)
- Exporting or deleting your account and all associated data
No Medical/Therapeutic Claims
Important: Wallies and its AI companions are NOT a mental health service, medical device, or therapy tool. AI companions do not replace professional healthcare providers. If you are in crisis, contact a licensed professional or your local emergency services.
Ongoing Improvement
We continuously monitor AI performance, user feedback, and emerging best practices to improve safety, reduce biases, and enhance your experience. Updates to AI models and policies are communicated transparently.