Saga Privacy Policy
Effective date: November 10, 2026. Owner: Saga is a product owned and operated by Circe Social Co.
1. Who we are and what this covers
Saga is a personal video product owned and operated by Circe Social Co. ("Saga," "we," "us"). You use Saga to upload personal photos and videos, have them edited into films by automated systems, and share those films. This policy explains what information we collect, how automated systems and our service providers process it, how we share and retain it, what we can and cannot delete, and the choices you have.
This policy covers the Saga mobile app and the Saga websites at saga.you, including public film pages that anyone can watch through a share link without an account.
• Legal contact / address: Circe Social Co., 1955 Courtland Ave, Oakland,
CA 94601.
• Jurisdictions offered: United States only.
• Privacy contact: privacy@saga.you.
• Support contact: support@saga.you.
• Changes: When we make a material change we will update the effective date
and, where required, notify you in the app or by email.
2. Information you provide
• Account and identity. Your email address and password, handled through AWS
Cognito; your profile display name, avatar, and bio.
• Collaboration. Project titles, descriptions, story/mood/tone/location
choices, edit prompts and steering text, comments, timestamp tags, and the email addresses you enter to invite collaborators.
• Media you upload. Original photos and videos — including their bytes,
filename, type, dimensions, duration, and any metadata embedded in the file (see Section 5 on embedded location) — plus uploaded music and voice audio, and voice reactions. We derive thumbnails, proxies, transcripts, and rendered films from this content.
• Guest contributions. If someone watches a shared film without an account
and leaves a named reaction, we collect that name and reaction as pseudonymous guest content visible to the project owner.
• Support, feedback, and reports. Feedback text, an optional screenshot, the
route/app-version/platform/viewport context attached to it, and abuse reports (target, category, and optional free-text detail).
• Payments. If you buy a subscription or film, checkout happens on
Stripe-hosted pages. Payment-card details are entered at Stripe and are not stored in Saga's systems; we receive a Stripe customer reference, subscription or entitlement status, and billing events.
3. Information collected automatically
• Identifiers. Internal account, project, asset, and job IDs; your Cognito
subject; a stable browser/device key; platform and model hints; and, after you grant permission, an Expo/APNs push token.
• Share-view network data. When someone opens a public film link, we record
the viewer's IP address, user agent, first-view time, and a persistent pseudonymous viewer identifier stored on the viewing device, which can span multiple share links, used for view counts and product analytics.
• Usage and telemetry. Upload, editing, approval/rejection, voting, comment,
and usage events; per-second watch telemetry (play, progress, skip, rewatch, drop-off, completion, playhead position, watched milliseconds, app version). When you delete your account, the link between this telemetry and your account is severed; the resulting de-identified playback facts are retained indefinitely for product analytics.
• Operational logs. Standard server, security, and error logs from our AWS
infrastructure, and app context (version, platform, route, locale) attached to feedback.
4. How we use information
We use information to authenticate you and run your account; to upload, process, edit, transcribe, and render your media; to power collaboration and sharing; to deliver notifications; to bill for paid features; to provide support; to secure the service and prevent abuse; to moderate content and comply with law; and to maintain, debug, and improve Saga.
Section 5 describes what is sent to each provider. We describe a provider's training practices only where we have verified them:
• Amazon Web Services. Saga's AWS account is opted out of AWS using your
content to improve its AI services, including Amazon Transcribe and Amazon Rekognition. This account-scoped setting does not cover other AWS accounts, and it does not mean the functional data used to provide a service is immediately deleted. Amazon Bedrock does not use your content to train AI models and does not share it with the companies that make those models.
• ElevenLabs. Saga has turned off ElevenLabs' option to use the content Saga
sends for model training. Under ElevenLabs' terms, this setting applies to content sent after it was turned off.
• Google Cloud. When Saga uses Google Cloud video analysis (Section 5), Google's
terms for that Vertex AI service state that Google does not use this content to train or fine-tune models without permission.
We do not make a blanket "we don't train on your content" claim for Saga itself while our diagnostic logs can still contain model responses. Otherwise we describe only protections we have verified, such as the abuse/CSAM detection disclosure in Section 5 and the verified-processor statements in Section 7.
Provider retention. Retention practices differ by provider and feature; Saga does not promise zero retention across all AI services. Providers may retain some inputs and outputs under their applicable service terms and settings for service functionality, security, abuse prevention, or troubleshooting. We assess these practices separately from whether content may be used for model training. Provider-specific disclosures describe the content shared and applicable retention and deletion practices, with verified limits identified where available. Deleting content from Saga does not necessarily remove provider copies retained under those disclosed terms, including applicable backup, security, or legal exceptions.
5. AI and automated processing
Saga's core function is automated. Your media and instructions are processed by automated systems and foundation models to analyze clips, transcribe speech, generate edit structure, and render films.
• Processors. Editing and analysis run on Amazon Bedrock (an Amazon
Web Services service), except video clip analysis, which may use Google Cloud (described below). Bedrock receives images and video frames, transcripts, editing instructions, and file details. Its configured foundation-model providers are OpenAI (GPT models) and Anthropic (Claude). Depending on the feature, speech transcription uses Amazon Transcribe or ElevenLabs. The selected service processes the full audio track, not only portions used in the finished film; audio tagged as music is not transcribed. Amazon Transcribe reads and processes audio from Saga-controlled AWS storage; its receipt of a storage reference does not mean it cannot access the audio. Voice notes and voice reactions may still use ElevenLabs. ElevenLabs receives audio directly at its own service, outside AWS/Bedrock. For AI-generated music and sound effects, ElevenLabs receives text prompts that may be derived from your editing requests and film content. Provider-side retention and cleanup must be assessed for each route; changing the transcription provider does not establish zero retention for other audio features.
• Saga safety review. Saga may send images and video frames to Amazon Rekognition to help identify potentially unsafe content. Authorized reviewers
may also review content submitted for publication, flagged media, and reports. Automated checks can make mistakes.
• Google Cloud video analysis. To choose which moments go into your film, Saga
may send your uploaded videos to Google Cloud Vertex AI (Gemini) as video with their sound: a reduced-size copy of the whole video, with its transcript and notes Saga works out about its sound and timing. Google's global service may process this information outside your region. Saga has turned off Google's optional data caching and request-and-response logging for this service. Google may retain requests flagged for potential abuse for up to 90 days, and authorized Google employees may review them for safety investigations. If a video can't be analyzed this way, Saga uses still frames and the transcript with Amazon Bedrock instead; in some of these cases Google has already received the video.
• Embedded location coordinates. Saga does not read your phone's live GPS
or Core Location. However, a photo or video you select can contain location coordinates embedded in the file (photo EXIF GPS, or a video container location tag). Our media pipeline extracts those coordinates and persists them server-side in media-metadata artifacts, and your original file (which also carries them) is retained in versioned storage. From the coordinates we derive a coarse place name using an offline lookup on our own servers — we do not call any location provider. That derived place name is stored server-side and linked to your account. Only that place name, never the coordinates, is sent to our AI providers, where it informs the edit. This is not live-phone tracking, and we do not use it for advertising. Saga v1 extracts and retains these coordinates rather than stripping them; a future stripping option may be offered but is not part of v1.
• Written descriptions of people. To assemble films, the AI analysis
generates and stores short written descriptions of the people in your clips — their appearance, clothing, and expressions — on a per-clip basis. Saga does not run face recognition or any biometric identification, and stores no faceprints or biometric templates.
• Support triage. Support feedback you submit is summarized by a foundation
model before it reaches our issue tracker.
• Accuracy. Automated outputs can be wrong, incomplete, or unexpected. You
should review films before relying on or sharing them.
• Automated abuse and CSAM detection. Separately from any product-improvement
use of content, the AI providers that process your media may run automated abuse and child-sexual-abuse-material (CSAM) detection as part of their service. Where a provider's system flags content, that provider may retain and human-review the flagged input and report apparent CSAM to the authorities (including NCMEC) under the provider's terms and applicable law. This is the provider's process, not a Saga retention promise, and it is not a blanket "never retained" statement. When Saga uses OpenAI models on Amazon Bedrock, AWS keeps requests that its automated classifiers flag for up to 30 days for automated abuse detection, and does not share them with OpenAI. We will state other providers' retention periods once they are verified.
• Retention/region/human review. Except for the abuse-detection basis and the
Google Cloud terms described above, and the ElevenLabs transcript deletion and backup terms described in Section 9, we have not yet established, and so do not claim, provider-side retention terms, data region, or routine human review. We will state these only once the relevant provider agreements are verified against the deployed configuration.
6. Sharing and collaboration
• Collaborators. People you add to a project may see that project's media and
content according to their role.
• Share links. A film share link is a bearer capability: anyone who has a
live link can watch the film, and can forward it. Revoking a link stops new resolutions but cannot recall copies already loaded. Some links can also let a viewer join the project; this is chosen when the link is minted.
• Guest data. Guest names and reactions are visible to the project owner.
• Notifications are disclosures to delivery providers. Push and email are
not content-free. Depending on the message, Expo/Apple (APNs) and AWS SES receive titles, choice prompts, synopsis/runtime copy, identifiers, and deep links; a film-ready message may include a short-lived presigned still-image URL (configured expiry ~900 seconds). URL expiry is not message deletion: the notification or email can remain at the provider, on your device, or in a recipient's mailbox, and we cannot retract copies already delivered.
7. Service providers and disclosures
We share data with providers who process it on our behalf:
Provider — Role — Data involved
Amazon Web Services — Identity (Cognito), media storage (S3), database (RDS), processing (Lambda/ECS/SQS), delivery (CloudFront), email (SES), AI (Bedrock — creative editing and support-feedback summarization; Transcribe; Rekognition safety scans), logging (CloudWatch) — Nearly all account, media, content, usage, and diagnostic data, including support feedback summarized before it reaches the issue tracker; Bedrock requests to OpenAI models flagged for potential abuse may be retained up to 30 days
Stripe — Hosted checkout/portal, subscriptions, entitlements — Email, Saga user reference, purchase/billing status; card data at Stripe
Expo — App updates and push transport — App/device metadata, push token, notification content
Apple — Photo-library permission, APNs delivery, App Store distribution — Selected media, device token, notification content
ElevenLabs — Feature-dependent speech transcription and generated music/sound effects; called outside AWS/Bedrock — Full audio for transcription; text prompts derived from editing requests and film content for generation; provider-side retention is not represented as universally zero
Google Cloud — Video clip analysis (Vertex AI Gemini), when enabled; called outside AWS — Reduced-size copies of uploaded videos with their sound, transcripts, and derived notes; requests flagged for potential abuse may be retained up to 90 days
GitHub — Optional private feedback sink — Feedback text, context, IDs, screenshot reference
Circe Social Co. is the controller of your personal information; the providers above act as our processors/service providers. We may also disclose information for legal or safety reasons, in a corporate transaction, or to professional advisers. Our infrastructure and AI providers (AWS, ElevenLabs) are contractually prohibited from selling or sharing your personal information. We are completing a review of every provider in our data map, including Google Cloud and our payment processor, before making a service-wide statement.
8. Safety, reports, and moderation
You can report content and accounts. We collect the category reported, your reporter details, and any evidence, and we preserve it as safety evidence. We investigate reports and may restrict, remove, or quarantine content or accounts, but we do not promise that every report results in removal, and we do not promise universal monitoring. Emergency and illegal-content escalation follows our internal takedown runbook. We review reports promptly. A logged-out report path is provided on every public film page before public sharing is enabled, so viewers without an account can report content.
9. Retention and deletion
We keep information for as long as needed to provide Saga and for the additional purposes described here. Deleting content is not instantaneous across every copy, and some copies persist on defined schedules:
• Active storage. Media lives in private, encrypted, versioned S3 with no
general expiration; database rows persist while your account and resources exist. Incomplete multipart uploads expire after 7 days.
• Deletion today. You can delete individual assets and projects, or your whole
account in Settings → Account → Delete account (or by contacting support@saga.you). Deleting your account signs you out and removes your login, uploaded media, projects and share links from Saga's active systems. Films already made in projects shared with others stay with those people. Some records may be kept where described below (for example safety evidence or legal holds). When ElevenLabs transcribes audio, Saga asks ElevenLabs to delete the transcript as soon as Saga receives it; ElevenLabs may keep deleted records in its backups for up to 30 days, and may keep limited debugging or safety data under its own terms. Backups, queues and logs on our side age out on the schedules below. We do not publish a fixed completion time and will not claim one we cannot prove in production.
• Secondary copies (facts from configuration, not a full schedule). Database
backups are retained ~14 days; deleting a live row does not selectively erase it from an existing backup. Processing queues and dead-letter queues retain message bodies ~14 days. Worker container logs are copied to CloudWatch and retained 30 days (an ECS log-group override); API and Lambda logs are retained 90 days; identity logs are retained 30–90 days depending on the log group. Worker logs may contain a short excerpt of the editing instructions you typed. Deleting your data does not individually purge these; they age out on schedule or require operator-level deletion.
• Legal holds and retained records. We may retain certain records (e.g. safety
evidence, billing records, or data under legal hold) after deletion where required.
• Viewer telemetry for public links. When people open your public film links,
we retain rows about those viewers — including IP address and user agent (see Section 3). Those viewers are not account holders, so these rows are not covered by account deletion and have no fixed expiry today. A retention lifecycle for these rows is a separate backend task, not yet scheduled.
• Delivered messages. We cannot recall notifications or emails already
delivered to a device or mailbox.
To request deletion or ask about the status of a request, contact support@saga.you.
10. Security
Your media is stored in private S3 buckets, encrypted in transit and at rest, served over SSL only, versioned, and blocked from public access; film stills are served through short-lived signed URLs. We use access controls and account security measures. No method of storage or transmission is perfectly secure, and we do not guarantee absolute security.
11. Your choices and rights
• Permissions. You control photo-library and notification permissions through
your device.
• Share links. You can revoke share links at any time (subject to the limits
in Section 6).
• Access, correction, deletion, and other rights. Depending on the US state
you live in, you may have rights to access, correct, delete, or obtain a copy of your personal information, to opt out of sale/sharing (we do not permit our verified providers to sell or share your personal information; see Section 7), and to appeal a decision — including the rights California residents have under the CCPA as amended. Contact privacy@saga.you to exercise them; we will not discriminate against you for doing so. Saga is offered in the United States only, so this policy does not provide EEA/UK legal bases or international-transfer terms.
12. Children
The media you upload may depict children, even though accounts are held by adults. You must be 18 or older to hold a Saga account, and Saga has no child-account mode in this version. We handle requests concerning child subjects and child safety as described in Sections 8 and 11.
13. International transfers
Saga is offered in the United States only and runs on US-based AWS and provider infrastructure, except that processing by Google Cloud's global Vertex AI service may take place outside the United States. We do not offer Saga outside the US at launch, so this policy does not provide an international-transfer mechanism. If that changes, we will update this section before offering Saga in other countries.
14. Contact
Privacy requests: privacy@saga.you. Support: support@saga.you. Postal/legal contact: Circe Social Co., 1955 Courtland Ave, Oakland, CA 94601.