PRIVACY.md

Rama main · réplica de lectura del repositorio original.

# Privacy Policy

**Effective date:** 2026-06-14
**Service:** Slate, a Discord application
**Operator:** whoever runs an instance. Skyphusion Labs ships the code and does **not** operate Slate as a service (self-host only).

This policy explains what data Slate processes, why, who it is shared with, and how long it is
kept. Slate is open-source software (AGPL-3.0); anyone may run their own instance, in which case
that operator is the data controller for their instance. This document is a reference policy
for operators; Skyphusion Labs does not run a public hosted instance of this product.

> The product-wide commitment this policy is written against is
> [`PRIVACY-COMMITMENT.md`](PRIVACY-COMMITMENT.md), a pointer to the canonical copy at the
> constellation hub. Slate is **self-host only**: Skyphusion Labs ships the code and does not run
> Slate as a service, so the operator of an instance holds its data and carries the controller's
> duties.

## What Slate is

Slate is an AI assistant that participates in a Discord channel to help a team plan and develop
films. It reads conversation in the channels it is added to, maintains a structured "storyboard
brief" in the background, generates images, searches the web, and submits projects to the Vivijure
render pipeline. To do this it uses the Discord **Message Content** intent, which means it can read
the text of messages in the channels where it is active.

## Data Slate processes

- **Message content** in channels Slate is configured to listen in, plus direct messages to the bot
  and messages that @mention it. This is used as the input to the AI model and to build the
  storyboard brief.
- **Discord identifiers and display names** as they appear in conversation (so Slate can address
  people and attribute lines), and **channel IDs** (used as the key for a project's stored state).
- **Image attachments** you post (mood boards, reference stills, concept art). These are fetched
  from Discord's CDN over HTTPS and passed to the AI model for the current turn only. They are
  **not** written to disk and **not** stored in the conversation history (only a text placeholder
  is stored).
- **Storyboard data** that Slate derives from the conversation: the storyboard brief, a rolling
  window of recent conversation history, a brief "undo" history, and pending render jobs.
- **Knowledge base entries** you add with `!learn` (text or fetched URL content), which are embedded
  and stored for later semantic recall.

## Where the data goes (subprocessors)

To provide the service, Slate sends data to the following third parties:

| Provider | What is sent | Why |
|----------|--------------|-----|
| **Anthropic** (Claude, via Cloudflare AI Gateway) | message content, attached images | generate the assistant's responses |
| **Cloudflare** (D1, Vectorize, AI Gateway, Browser Rendering) | storyboard state, knowledge entries, fetched pages | storage, embeddings, model routing, headless page fetches |
| **Brave Search** and **Tavily** | search queries the model chooses to run | web search and research |
| **skyphusion-llm-public** | image generation prompts | character portraits and scene thumbnails |
| **Vivijure API** | storyboard bundle, cast data | render submission and cast sync |

If the operator runs Slate in its **ollama fallback** mode (no Cloudflare AI Gateway token), message
content is sent to a self-hosted model instead of Anthropic, and image attachments are reduced to a
text placeholder.

Slate does **not** sell your data or use it for advertising.

## Storage and retention

- Project state (brief, recent history, undo stack, render jobs) is stored in **Cloudflare D1**,
  scoped per Discord channel. It persists so the project survives a restart.
- Knowledge base entries are stored in **Cloudflare Vectorize** until removed.
- `!reset` clears the calling channel's project and conversation state.
- To request deletion of a channel's stored data or knowledge entries, contact the operator (below).

## Data scoping and security

- D1 session data is scoped per Discord channel ID; Slate does not read another channel's project.
- Secrets and API tokens are never stored in conversation data. See
  [SECURITY.md](SECURITY.md) for the security design and how to report a vulnerability.

## Children

Slate is not directed to children. You must meet Discord's minimum age (at least 13, or older where
your jurisdiction requires) to use Discord and therefore Slate.

## Changes to this policy

We may update this policy as Slate evolves. Material changes will be noted in
[CHANGELOG.md](CHANGELOG.md) and reflected in the effective date above.

## Contact

Questions or data-deletion requests: **conrad@skyphusion.org**, subject
`[PRIVACY] skyphusion-slate`.