Privacy, without hand-waving.

01 Scope and roles

Fracktron is an independently operated broadcast laboratory and community system. It receives information in three broad ways: information you send directly, information provided by a linked platform, and technical information produced when a browser, bot, API client, or broadcast tool uses the service.

• Direct input: chat messages, commands, uploads, voice samples, forms, and account settings.
• Platform data: account IDs, display names, avatars, channel or server context, and authorization grants.
• Operational data: requests, errors, playback events, stream state, and abuse-control signals.
• Derived data: memories, embeddings, rankings, summaries, aggregates, and generated media.

Twitch, Discord, Kick, and other platforms independently process activity under their own rules. Fracktron does not control copies those platforms or community members retain.

02 Shared session and OAuth

Public pages open without an account. Where a feature requires sign-in, Fracktron uses the auth service's host-only session cookie across same-origin apps under fracktron.com. Studio and account tools ask that auth service for identity instead of creating separate browser accounts.

The session begins through Twitch OAuth. Fracktron stores the server-side session record and basic Twitch profile fields needed to identify the account, such as the platform user ID, login, display name, and avatar URL. The browser cookie is marked Secure, HttpOnly, and SameSite=Lax. It is intentionally long-lived, and the server-side session does not expire by default. It ends through explicit logout, revocation, or account deletion.

The dashboard can link additional providers for optional features. Depending on what you connect, Fracktron may store provider IDs, profile fields, granted scopes, access tokens, refresh tokens, and token status for Discord, Kick, Google, X, Spotify, or other displayed integrations. Provider tokens stay server-side; they are used only for the linked feature and can be revoked from Fracktron or the provider.

03 Public pages and traffic

The homepage, Studio directory and public viewers, TrackTron, command reference, API documentation, public games, and listed experiments can be opened without signing in. Public channel names, stream facts, category history, quotes used by a game, and other community material shown there should not be treated as private.

Requests and audience estimates

Like any web service, Fracktron's servers receive network information needed to answer a request, including an IP address, user-agent information, requested route, and timestamp. Ordinary access and security logs may temporarily contain request details.

The reporting pipeline normalizes public routes, removes query strings, and does not retain raw IP addresses, cookies, request headers, or full user-agent strings in its analytics database. Approximate visitor counts use a rotating pseudonymous value and are audience estimates, not cross-site identity profiles. Sensitive overlay routes are excluded from traffic reports.

The static homepage does not load advertising tags, Google Fonts, or a third-party browser analytics SDK. Following an external GitHub, Twitch, Discord, or Egg link causes that destination to receive a normal web request under its own policy.

04 Chat, commands, and memory

When Fracktron is present in a channel, server, direct-message flow, Studio chat, or enrolled overlay community, it can process message text, sender and channel identifiers, display names, timestamps, reply context, command arguments, links, attachments, and its own outgoing responses. Channel owners may explicitly consent to chat logging when enrolling an overlay.

That information supports replies, moderation, rate limits, saved quotes, conversation continuity, search, evaluation, and the multi-surface brain. Fracktron can derive summaries, topics, relationships, embeddings, and memories from prior conversation so later interactions are not stateless. A direct message is delivered less publicly than channel chat, but it is still processed and may be stored by Fracktron and the platform.

Chat history is intentionally a long-term input to memory and social features. It is not covered by the general automatic cleanup that removes some short-lived operational records. Avoid asking the bot to remember information that should not persist.

Private Messenger

The separate Messages service encrypts message bodies and room keys in participating clients. Synapse and the Messages gateway handle ciphertext plus delivery metadata such as account, device, room and community membership, relationship state, timing, and approximate message size. Private Messenger content does not enter Fracktron chat memory, AI prompts, content filters, or notification previews.

Every active Messenger enrollment is listed to other signed-in Messenger members by default with its permanent Messenger handle and current Twitch login, display name, and avatar. Members can hide from directory browsing in People settings, but someone who already knows the exact Messenger handle can still send a friend request. Friend requests, accepted friendships, blocks, directory preference, and community membership are stored server-side. Blocks stop new direct rooms and invitations but cannot erase messages or encryption keys already received, and shared-community rosters remain visible.

05 Features that publish or reuse content

Confessions

A supported confession is submitted through a private Discord conversation, rate-limited, queued, and published after processing. The published confession omits the submitter's identity. Fracktron and Discord can still process the sender's account and service metadata to receive the message, enforce abuse controls, and operate delivery.

Confession text or attached media can be relayed to participating Discord communities and to overlays whose operators opted in. Recipients can copy or repost it. This is pseudonymous publishing, not end-to-end anonymity from the service, platform, or recipient communities.

Saved quotes

The save command stores the message being replied to, its attribution and community context, and cross-posts it to the configured community quote channel. Saved quotes can later appear in search, bot replies, statistics, or the Who Said That? game. Do not save content that should not become a durable community artifact.

Build Log

Direct text posts by an explicitly configured author in the designated Discord changelog channel are copied to the public Build Log with their publication time. Discord message, channel, server, and author identifiers remain internal. Edits update the public entry and deletions unpublish it; replies, thread posts, bot or webhook posts, and attachment-only messages are excluded.

Global image library

Images submitted through content commands may be stored with contributor and moderation metadata. Adds are global by default unless the command explicitly uses a local scope, so a global item can be searched or reused across participating communities and Fracktron surfaces. Generated and curated images may enter the same library.

Voice samples and generated speech

Voice-cloning commands can store an uploaded audio sample, a voice reference, ownership metadata, and derived audio needed to generate speech. A stored voice may be listed and used where voice commands are enabled. Upload only your own voice or a sample you are authorized to provide. Voice removal is owner- or moderator-controlled rather than an anonymous automatic process.

06 Studio diagnostics and broadcast telemetry

Studio processes broadcast video and audio so viewers can receive a live stream. It also handles channel identity, chat, viewer or broadcaster permissions, selected quality and delivery path, and state needed to reconnect a session.

Playback diagnostics can include stream and page identifiers, a pseudonymous viewer key or signed-in viewer identity, quality selection, player state, errors, timing and buffering observations, WebRTC or HLS statistics, browser and device characteristics, IP address, user agent, and timestamps. Publisher diagnostics can include ingest state, protocol, media health, connection events, and technical reports submitted by a broadcaster.

Raw diagnostic details are used to investigate broken playback, publishing, browser layout, and abuse. Direct network and browser identifiers are kept for a bounded diagnostic window and then removed from raw player events; summarized session and service-health records can remain longer. Restricted diagnostics are available only to the relevant broadcaster or operators, while public viewers receive coarse status.

Overlay and chat-reader URLs contain bearer tokens. Anyone with one can access the capability it represents. Keep those HTTPS URLs out of screenshots, chat, source control, and public browser history; rotate the URL from the dashboard if it leaks.

07 Third parties and model services

Fracktron shares data only as needed to operate a selected feature, communicate with a platform, provide infrastructure, or respond to a valid legal or security requirement. Relevant recipients can include:

• OAuth and community platforms such as Twitch, Discord, Kick, Google, X, and Spotify when you sign in, link an account, request platform data, or publish through them.
• Broadcast and media services involved in stream delivery, embeds, clips, music metadata, or user-requested media retrieval.
• Hosting, network, backup, and error-diagnostic services used to keep Fracktron available and recoverable.
• AI, search, speech, image, and other tool providers when a configured feature needs them. Relevant prompts, conversation context, media, or search terms may be sent to that provider.
• Public repositories and destinations such as GitHub when you choose to follow a link or participate there.

Some model and media processing is self-hosted and some can use external services. The mix may change as experiments change, so the safest rule is not to send material that an operational provider should never receive.

08 Retention and backups

There is no single retention period for every Fracktron surface. Data is kept according to its purpose:

• Browser sessions and linked-provider records remain until logout, revocation, account deletion, or an operator removes them.
• Short-lived errors, raw speech transcripts, detailed playback events, denylist entries, and quota counters are subject to scheduled pruning.
• Normalized traffic events, aggregate tracker observations, session summaries, and service-health history can be kept longer for trends and reliability work.
• Chat history, memories, saved quotes, content-library items, voice references, community configuration, and moderation records may be retained long-term because they power ongoing features.
• Backups rotate separately and can retain a deleted record after it disappears from the live service until the relevant backup ages out.

Public posts, cross-posted quotes, confessions, images, generated media, or messages copied by other people may persist outside Fracktron even after the original record is removed.

09 Security

Fracktron uses HTTPS, server-side sessions, restricted account controls, scoped authorization checks, token revocation, protected diagnostics, and operational monitoring. Sensitive browser credentials are not intentionally placed in public page content, and the shared session cookie is not readable by page JavaScript.

Some feature URLs, API keys, and grants are credentials in their own right. Keep overlay URLs and API keys private, use logout on shared devices, revoke connections you no longer use, and use the account-switch control rather than sharing a login.

No internet service can promise absolute security. Experiments can change, software can fail, and authorized recipients can misuse data. Report a suspected leak or account problem promptly using the contact path below and do not publish the secret in a public issue.

10 Your choices, access, and deletion

• Use public pages without signing in.
• Log out, switch Twitch accounts, or revoke linked provider grants from the account controls and provider settings.
• Disable enrolled overlay or Discord features and rotate a leaked overlay URL.
• Hide an enrolled Messenger profile from directory browsing and manage friend requests or blocks from People settings.
• Use documented owner or moderator commands to remove eligible images, voices, or other community content.
• Ask for a review, correction, export where practical, or deletion of account-associated Fracktron data.

Broader access and deletion requests are handled manually; Fracktron does not promise an unsupported automatic account-deletion workflow. A request may require proof that you control the relevant platform account and enough detail to locate the records.

Removal may be limited where data belongs to another account or community, is needed for security or abuse records, has been aggregated without a practical identity link, remains in rotating backups, or was already published or copied to another platform. The operator will explain what can and cannot be removed for a specific request.

11 Contact