Privacy Policy

Updated August 16, 2026 · Beta policy

Sprigkin turns story seeds into branching stories, growing plants, rough storyboard sketches, and finished comics, alongside an optional companion and user-controlled memory. This policy explains the data needed to make that experience work and the controls available to you.

Who operates Sprigkin

Sprigkin is a service of Cluntra Inc. In this policy, “Cluntra,” “we,” and “us” refer to Cluntra Inc. Contact us at contact@cluntra.com.

Information we process

• Account details such as your email address, authentication provider, account identifier, and support messages.
• Feedback you submit, the product page and bounded feature context attached to it, an optional contact email, and an optional screenshot. Feedback context describes one current product output and does not include a raw chat history.
• Your chosen display name, current focus, profile notes, chat messages, memory cards, ideas, drafts, creator preferences, pet progress, and other content you choose to add.
• For a Guided Story Run, your seed, selected story beats, user-written directions and final beat, story canon, current unselected option pair, plant choice and growth events, visual-continuity notes, rough sketches, final comic, and limited generation and storage metadata.
• If private image creation is enabled and you use it, the confirmed Idea context and art direction you submit, three generated rough drafts, an optional polished final selected from those drafts, and limited Creation metadata such as format, dimensions, and timestamps.
• Images you choose to attach to a chat, plus limited file metadata such as name, type, and size.
• Operational information such as IP address, browser and device data, timestamps, security events, error logs, product actions, and campaign tags included in a link you followed.
• Subscription status, renewal dates, purchase channel, and payment provider identifiers. We do not store full payment-card details.

Why we use it

We process this information to create and secure your account; save and personalize your little world; provide chat, memory, idea-garden, and creator features; process subscriptions; answer support requests; prevent abuse; diagnose reliability problems; comply with law; and understand and improve the service. Where a legal basis is required, we rely on performing our agreement with you, our legitimate interests in operating a safe service, consent where requested, and legal obligations.

How AI and memory work

Sprigkin keeps the cards you can inspect, edit, export, and delete in its product database. When long-range recall is available for your account, conversation turns are also sent to EverMind's EverOS service to build semantic recall. EverOS is a supporting recall layer, not the sole record of your visible memories.

The profile you write during onboarding is stored with your editable Sprigkin world and is used as direct context for replies. When long-range recall is available, creating or updating that profile also sends a user-confirmed profile message to EverOS. Authentication credentials and your account email address are not included in that profile payload.

When you attach an image, the image is sent to the configured visual model for that reply. Sprigkin keeps only limited attachment metadata in your conversation history; it does not write the image bytes into your saved transcript or EverOS memory.

Web search starts in Auto mode. Broad-knowledge, how-to, and current questions may send the text of that question to Moonshot AI's Kimi web search service. You can force search for one reply or turn web search Off before sending. The isolated search request does not include your profile, saved memories, account email, or earlier chat history.

DeepSeek ordinarily processes companion text requests. If the configured primary text model is temporarily unavailable, Sprigkin may use Moonshot AI/Kimi as a bounded backup for that reply. The model receives the reply context needed for the feature, which can include your profile, visible memories, recent conversation, relevant EverOS recall, and research for the current turn. This backup does not turn on web search: choosing Search Off still prevents a web lookup.

When you ask the Idea Garden to suggest a next direction, Sprigkin sends the configured text model a bounded copy of that Idea: its original seed, the state and choices you confirmed, recent progress, and fingerprints of directions you declined. A Daily Incubation request may also include up to three active memory clues that the app selected as related to that Idea. It does not include your account email, full profile, unrelated memories, or general chat history. Model proposals remain suggestions until you explicitly choose, edit, or replace one.

The story-first Idea Garden uses a separate Guided Run record. For each round, the configured text model receives the story seed, language, the canon you explicitly selected or wrote, continuity notes, the current round, and your chosen companion's name. It does not receive your account email, profile notes, general chat history, unrelated Ideas, or personal Memory. Two proposed paths remain non-canon and are erased or superseded unless you select one. Regenerating paths, asking for an idea, or saving your place does not add a story fact or teach personal Memory. The final beat is written by you; Sprigkin does not generate it for you.

Sprigkin does not currently send product analytics to a separate third-party analytics service. For Guided Stories, limited product-action events may be kept with your Sprigkin state, such as opaque Idea and Run identifiers, the plant species you chose, plant-stage milestones, and timestamps. Those event properties do not contain your story seed, option text, selected beat text, canon, final beat, fictional characters, events, dialogue, route choices, or image prompts. The story itself remains in its separate, owner-scoped Idea and Guided Run records.

After you select a story path, Sprigkin may send an English visual prompt derived from that selected beat and its continuity notes to OpenAI's GPT Image 2 service for one loose storyboard sketch. Up to 20 selected beats may have sketches in one run. A failed sketch does not undo the selected story beat or plant growth. After you save your final beat, you may ask Sprigkin to plan and render a six-panel final comic. Generated story image bytes are stored in private Cloudflare R2 and owner-scoped metadata is stored in D1; they are not written to AppState, personal Memory, or EverOS. If a final comic fails, the completed story remains saved.

Movie seed help reads a bounded response from the official Wikidata Query Service on the server. It requests only an opaque film item URI, release date, release-place qualifier, and genre identifier for recently released films in a coarse server-owned country mapping derived from your market that have an English Wikipedia article; it does not request a movie label, title, or synopsis. Your original market text is not sent to Wikidata. The supported mappings are the United States, Canada, United Kingdom, Australia, and Mexico; other market values use the United States mapping. The suggestion keeps the retrieval time and a fixed Wikidata source link, not the selected item identifier. Server-owned allowlists convert the country mapping and recognized genre identifiers into the constrained query and broad genre and theme signals. The configured text model receives only those broad signals and the release date—not an item URI, genre identifier, raw query, title, synopsis, or source URL—to request an original, spoiler-free seed. The helper fails rather than claiming a movie is current without a verified source. Story and Life helpers do not query Wikidata.

Creative Taste contains only preferences you explicitly add or confirm. You can view, edit, or delete each preference. Creative Taste does not become a general memory card, enter EverOS, teach the pet from a single choice, or add Bond or room-growth points.

When Creation Studio reports Visualize as available and you request a Visual Set, Sprigkin sends the original seed, current accepted direction, confirmed decisions, explicit Creative Taste, and your art direction to OpenAI's GPT Image 2 service. Pending proposals, unrelated memories, account email, and general chat history are not included. Generated image bytes are kept in private Cloudflare R2 object storage, while D1 stores only ownership and technical metadata; image bytes and private object keys are never put in AppState, general Memory, or EverOS. A failed provider or storage attempt does not change the Idea or Creative Taste. Sprigkin first requests three rough composition drafts. One successful Visual Set allowance is recorded only after all three drafts are stored privately with their metadata. Failed or partial sets do not consume that allowance. If you choose a draft, Sprigkin can send that private draft back to OpenAI to create one included polished final; the final does not consume another set allowance. Visualize does not offer video generation. Sprigkin reports Visualize as available only when the creation schema, private R2 storage, image provider, account access, and remaining set quota all permit a new Visual Set.

Private Story and Creation images are not distributed through public or pre-signed sharing links. The app loads them from same-origin Sprigkin routes using your current bearer session. Each request checks that session and resolves the asset under the authenticated account; another account cannot use the asset path to load an image or an authenticated deletion request to delete it. The path itself has no separate expiry time or access power: access ends when the account session expires or becomes invalid, or when account deletion starts. Image responses are marked private and no-store.

When neural voice is available and you use the microphone button, Sprigkin holds that bounded recording in memory only long enough to send it to the configured cloud voice provider for speech-to-text processing, then discards the audio. When enabled, Alibaba Cloud Model Studio in Singapore is the primary turn-based transcription and streamed neural speech provider; a configured OpenAI voice service may be used as a fallback, and Safari audio formats unsupported by the primary provider may go directly to that fallback. The hidden transcript joins the same chat and may be sent to EverOS for long-range recall under your memory controls. Sprigkin sends the pet's reply text to the configured voice provider to create an AI-generated voice stream, but does not retain a separate recording of that reply. If cloud voice is unavailable, Sprigkin falls back to your browser's built-in speech recognition and synthesis; the browser or operating-system provider may process audio under its own privacy terms. Voice Mode does not display what you said, and Sprigkin does not save raw microphone audio to its app database or files in either mode. Providers may handle request data under their own retention and privacy terms.

Product feedback is a separate support workflow. It is not sent through memory extraction and does not teach the pet, create a memory card, or change Bond, Growth, or Idea state. Before saving feedback, Sprigkin removes page query parameters and fragments, accepts only a small set of contextual fields, rejects transcript/history fields, and redacts common credential and token formats. Resend receives the feedback so it can notify our product team. If you attach a screenshot, it is limited to a small PNG, JPEG, or WebP file, held in the notification outbox only until Resend acknowledges the email, and then erased from Sprigkin's database. If notification delivery remains unavailable, Sprigkin erases the pending screenshot after seven days and continues retrying the text-only support notification.

Service providers

• Cloudflare provides hosting, application storage, network security, and Turnstile abuse protection.
• Supabase provides account authentication, and Resend may deliver authentication and other transactional email through that setup.
• EverMind/EverOS provides long-range memory and recall.
• DeepSeek processes primary text-model requests, while Moonshot AI/Kimi may process image requests, isolated web-search questions, and bounded backup replies as described above.
• Alibaba Cloud Model Studio may process bounded microphone recordings and reply text in its Singapore deployment when optional cloud voice is configured. OpenAI may process bounded microphone recordings and reply text as a configured voice fallback, and the bounded confirmed Idea context used for a requested private still image when those optional features are configured.
• Stripe handles Web checkout, billing details, invoices, and subscription management. If an iOS subscription is offered, Apple handles the App Store purchase and payment record.

These providers process information on our behalf or as an independent platform where applicable, under their own terms and privacy commitments. Feature availability and the provider used can depend on configuration.

Sharing and retention

We do not sell personal information or share it for cross-context behavioral advertising. We disclose information to the providers above, when you direct us to share something, to protect people and the service, when law requires it, or as part of a business transaction subject to appropriate safeguards. Public pet cards contain only the summaries displayed in the share preview; raw chat and source text are excluded.

We retain account and product data while your account is active. When you delete a memory or account, we remove or queue removal from active product systems and configured memory providers. Limited billing, fraud, security, dispute, backup, and legal records may remain for as long as reasonably necessary or legally required. Retention can vary by provider and record type.

Feedback records remain available to handle the report while your account is active and are removed with account deletion, subject to any limited legal or security retention described above. A screenshot may remain for up to seven days while its support notification is pending; after a successful notification, Sprigkin erases the screenshot bytes sooner even while retaining the text feedback record.

Private Visual Sets and text works are independent Creations. If you delete only their source Idea, Sprigkin keeps the Creations by default, detaches the live Idea link, and retains the frozen source snapshot so you can still view the work or finish its included polished render. You can instead explicitly choose to delete the Idea and its Creations together. Deleting an independent Creation immediately makes its work and images unavailable, tombstones its metadata, and queues every related private R2 object for removal. After those objects drain, Sprigkin removes the work's title, content, source snapshot, set, job, and asset metadata while retaining only a content-free quota fence. Account deletion queues and removes all Creation objects and metadata; limited legal, security, or backup retention may still apply as described above.

Guided Runs use a separate deletion path. Rewriting an unfinished round removes the later canon and queues its later rough sketches for deletion. Deleting an unfinished Story Idea removes its Guided Run and private story images. A completed final comic can remain in Creations after its source Idea is deleted; explicitly deleting the completed run removes that comic too. Account deletion queues all Guided Run images for removal and removes the related run, canon, job, and comic metadata. A content-free lifetime allowance fence may remain until account deletion so deleting and recreating a run cannot reset the free-use limit.

Your choices and privacy rights

Settings lets you update your profile, edit or delete individual memories, export your data, and request account deletion. Creation Studio lets you view, edit, or delete Creative Taste. When Visualize is available to your account, authenticated Creation Studio controls let you view each draft and final and delete the Visual Set as one Creation; individual drafts are not separate deletion units. Search controls let you turn web search Off for a reply. Account deletion clears product data and asks configured memory providers to delete the associated user scope. If a subscription is active, Settings shows how to manage it and also lets you explicitly confirm immediate account deletion. Deleting the Sprigkin account does not automatically cancel an Apple-billed subscription, so Apple may continue billing until you cancel it with Apple. Some billing records may be retained where law requires it.

Fiction written in a Guided Run is not added to personal Memory, EverOS, or Guided Story product-action event properties. You can save and resume an unfinished run, revisit selected rounds, rewrite an unfinished branch after confirming that later pages will be removed, and access a completed final comic from Creations.

Depending on where you live, you may also request access, correction, deletion, portability, restriction, or objection, and may appeal a decision or complain to a regulator. We will not discriminate against you for exercising a privacy right. Email contact@cluntra.com from your account address so we can verify and respond to the request.

Storage, security, and international processing

Sprigkin uses essential browser storage and authentication technologies to keep you signed in and operate the service. We do not currently use advertising cookies. We use reasonable technical and organizational safeguards, but no online service can promise absolute security. Our providers may process information in the United States and other places where they operate, subject to appropriate protections required by law.

Age, changes, and contact

Sprigkin is intended for people aged 18 or older. We may update this policy as the product, providers, and law change. We will post the revised date here and provide additional notice when a change is material.

Privacy or support questions can be sent to contact@cluntra.com. This policy is a plain-English beta policy and has not been represented as reviewed by an attorney.