Wilds.ai Subprocessors
Effective Date: November 19, 2026
Previous version: November 16, 2026
This page lists every external service ("subprocessor") that wilds.ai may share user data with in order to operate the platform. We update it whenever a subprocessor is added or removed. For privacy questions or to exercise your data-subject rights, contact team@wilds.ai.
What this page covers
A subprocessor is a third-party vendor that processes personal data on our behalf. We list every vendor that may receive any of: account identifiers (email, account ID), prompt or response text, image or audio inputs, billing identifiers, telemetry, or content you have published. Vendors that only serve us static media or aggregated metrics with no user-identifiable payload are not listed.
We do not sell your raw prompts, transcripts, or character conversations to any of the vendors below.
AI Providers
These vendors generate or transform content (text, image, video, music, voice, 3D models, embeddings) on our behalf. The selection of a specific provider for a given request depends on your model-quality choice, the content rating of the character or world, and the content type. The film studio sends generation briefs and their selected look or composer settings to the provider for that kind; a still may also send a reference image, and a narration take sends the selected line and voice.
| Vendor | Purpose | Data sent | Region |
|---|---|---|---|
| Anthropic | LLM inference for the Claude models that power character chat, narrator prose, and most analytical tasks. | Prompt + response text, conversation history, system prompts; images sent for judging when a judge falls back to this vendor. | Stored in the US; processed in the US, Europe, Asia and Australia |
| OpenAI | LLM inference for the GPT-5.5 family (gpt-5.5, gpt-5.5-mini, gpt-5.5-nano) used in compiler runs, judges, and structured-output tasks; image screening of uploaded images; description of images uploaded in chat and for outfits; image generation (gpt-image-1.5); fallback text-to-speech. | Prompt + response text, conversation history, system prompts; uploaded images sent for screening; images uploaded in chat and for outfits, sent for description; images sent for judging; image prompts and optional reference images; text segments to vocalize. | US |
| OpenRouter | Gateway routing for mature-rated fiction inference (Llama 3.3, etc.) and as a refusal-retry fallback. | Prompt + response text; images sent for judging when a judge falls back to this vendor. | US/EU (provider-dependent) |
| Google Gemini API | LLM inference for narration regeneration, refusal fallbacks and judges; screen vision for Tagalong. | Prompt + response text, conversation history, system prompts; Tagalong screen frames; images sent for judging. | Provider-operated |
| fal.ai | Primary image generation (character avatars, scene art, sprites), video generation and music generation. | Image, video and music prompts, optional reference images. | US and other countries |
| Replicate | Image, video and music generation (character avatars, scene art, asset-forge outputs); MusicGen music beds in the film studio. | Image, video and music prompts, optional reference images. | Provider-operated |
| Stability AI | Alternate image generation (when configured); Stable Audio ambience generation in the film studio. | Image prompts, optional reference images; ambience briefs and generation settings. | US |
| Black Forest Labs | FLUX image generation when the image-provider chain selects BFL. | Image prompts, selected look settings and optional reference images. | Provider-operated; deployment region UNVERIFIED |
| ElevenLabs | Sound-effect generation, including film-studio sound effects; premium text-to-speech and NPC voice generation when enabled. | Sound-effect prompts; text segments to vocalize, voice profile selection when enabled. | US |
| Deepgram | Speech-to-text for voice input and default text-to-speech (Aura) for narrator and character voices, including film-studio narration takes. Deepgram's Model Improvement Program lets Deepgram keep request data for future model training unless a request opts out; wilds.ai opts every speech-to-text request out (mip_opt_out). Text-to-speech requests do not carry the opt-out. | Audio samples from voice sessions, text segments to vocalize. | US |
| Cartesia | Text-to-speech for narrator and character voices, when selected. | Text segments to vocalize, voice profile selection. | Provider-operated |
| Hume | Text-to-speech (Octave) for narrator and character voices, when selected. | Text segments to vocalize, voice profile selection. | Provider-operated |
| Simli | Live talking avatars for characters and Tagalong, at the Safe and Standard ratings only. Simli's Data Processing Agreement (s.1(c)) authorizes Simli to use the content and output it processes for training its AI and Vision Models. | Character portrait images, character voice audio. | Provider-operated |
| Meshy | 3D character and model generation, rigging and animation. | Character descriptions and 3D prompts, optional reference images. | Provider-operated |
Search & Web Retrieval
These vendors return external information used by deep-research, news search, and the verify-claims tool. Queries may include text you typed if you asked the agent to look something up.
| Vendor | Purpose | Data sent | Region |
|---|---|---|---|
| Firecrawl | Page scraping for deep-research + agent web tools. | Search query text, target URLs. | US |
| Serper | Google Search API proxy. | Search query text. | US |
| Tavily | AI-optimized search. | Search query text. | US |
| Brave Search | Independent web index. | Search query text. | US |
| NewsAPI | Recent-news search. | Search query text. | US |
We rank search results by relevance on our own servers, and an AI provider listed above may then order the best of them. Until November 11, 2026, Cohere ranked them and received the search query and the text of the results; until November 9, 2026, while a memory-ranking setting on our servers was on, Cohere also received the stored memories from your character conversations and your current message.
Stock Media Providers
We use these providers to source CC0 / royalty-free media for asset packs and inline content. We send only short, generic queries (no chat content, no account identifiers) and consume the public response.
| Vendor | Purpose | Data sent |
|---|---|---|
| Pexels | Stock photos and video. | Generic search keywords. |
| Pixabay | Stock photos, illustrations, video. | Generic search keywords. |
| Unsplash | Stock photos. | Generic search keywords. |
Infrastructure
| Vendor | Purpose | Data stored | Region |
|---|---|---|---|
| Linode (Akamai) | Application hosting, primary database, multiplayer relay, voice chat (self-hosted LiveKit), and Linode's daily backups of the server's disk, none kept longer than 14 days. | All canonical user data: account records, content, sessions, messages, memory. | Atlanta, GA |
| Linode Object Storage | Avatar images, generated art, audio assets, exports. | User-uploaded and generated media. | us-southeast-1 (Atlanta) |
| Backblaze B2 | Off-site backups of the primary database, encrypted on our server before upload with a key Backblaze does not hold, and at rest by Backblaze; copies sent before November 12, 2026, when that began, are encrypted at rest only. Each is deleted within 31 days; the newest stays until a newer one replaces it. | Database backups holding all canonical user data. | US-East |
| Cloudflare | DNS + CDN at the edge. | Standard request metadata (IP, user-agent, headers); no body inspection. | Global edge |
Authentication & Identity
These vendors are involved when you sign in with a third-party provider or pair an external account.
| Vendor | Purpose | Data exchanged |
|---|---|---|
| OAuth sign-in and identity verification. | Email, basic profile, OAuth tokens. | |
| Discord | OAuth sign-in and bot-pairing for the Discord channel. | Username, basic profile, OAuth tokens. |
| GitHub | OAuth sign-in. | Username, email (when public), OAuth tokens. |
| Microsoft | OAuth sign-in (when enabled). | Email, basic profile, OAuth tokens. |
Payments & Email
| Vendor | Purpose | Data sent | Region |
|---|---|---|---|
| Card processor | Card payments are paused (since June 10, 2026). The processor will be listed here before the first charge. Stripe, Inc. processed card payments until June 10, 2026 and keeps its own records of those transactions; we keep the related customer IDs and purchase history for accounting. | Historical: email, customer ID, billing address, partial card metadata, transaction history. We do not store full card numbers. | US |
| Resend | Email — account verification, password reset, account-deletion confirmation, payment receipts, policy notices, the messages you send from the "Write to us" form on the Contact page, and the notification emails you can turn off (digest, comments, milestones, messages, referrals, Creator Jams, low balance). Resend tells us when a mail bounces or is marked as spam. | Email address, message contents (no marketing mail); for the Contact page form, the name, email address, topic and message you enter. | Stored in the US |
Analytics
Optional analytics. PostHog, Google Analytics and Microsoft Clarity load only after you opt in through the cookie banner or settings; until then your browser sends them nothing. Our servers send them your purchase and gameplay events only under the same yes, saved to your account when you are signed in. Plausible sets no cookies. While your browser sends Global Privacy Control, Wilds treats it as a refusal of analytics consent.
| Vendor | Purpose | Data sent | Region |
|---|---|---|---|
| PostHog | Product analytics — pageviews, named events, funnels, masked session recordings. Consent-gated: receives nothing until you opt in. | Named events, an account or guest ID, masked DOM text. | US (EU residency available on request). |
| Google Analytics 4 (Google) | Pageview and event analytics. Consent-gated: receives nothing until you opt in. | Pageview and event metadata, an account or guest ID, device and browser data. | Provider-operated |
| Microsoft Clarity | Heatmaps and session recordings. Consent-gated: receives nothing until you opt in. | Page interactions and session recordings, device and browser data. | Provider-operated |
| Plausible | Lightweight, cookieless pageview analytics. | Anonymous pageview metadata. | Self-hosted on our own server |
Communication Channels
These are surfaces wilds.ai integrates with at the user's explicit pairing.
| Vendor | Purpose | Data exchanged |
|---|---|---|
| Discord | Optional bot-paired notification channel. | Messages you send to the bot, reactions you react with. |
| Telegram | Optional bot-paired notification channel. | Messages you send to the bot. |
| Twitch | Optional stream-chat connection for Tagalong. | Account tokens, your channel ID, live chat messages on your channel. |
| YouTube | Optional stream-chat connection for Tagalong. | Account tokens, your channel ID, live chat messages on your channel. |
Staff Notifications
Internal notices that staff use to answer support requests and to triage accounts, abuse, and safety reports. They go to staff-only channels, and other users cannot see them.
| Vendor | Purpose | Data sent |
|---|---|---|
| Discord | Staff-only audit channel on the wilds.ai Discord server. | Account events (display name, public profile link if a username is claimed, masked email, account ID, sign-in or verification method, linked Discord account); support requests opened in wilds.ai or through Wilo (a notice that a request is waiting, which carries its reference, how long it has waited and a link to it on our staff pages, and no message text, name, email or ID); content and comment reports (the reporter's display name and profile link if a username is claimed, what was reported and its ID, category, description or reason, report ID); referral-abuse alerts (the referrer's username or masked email, account ID, referral counts and times). The privacy policy's "Operational audit notifications" item lists every field. |
Adding or removing subprocessors
We update this page each time a subprocessor is added or removed. If you operate under a Data Processing Addendum and need email notice of changes, contact team@wilds.ai to be added to the notification list.