The Witness

Privacy Policy

Effective November 4, 2026

1. Who operates your instance

This policy describes the official The Witness bot and thewitness.dev, operated by Steve (Straconis). For self-hosted deployments, the person or organization running the instance controls its processing and storage. Using the published source does not itself send your self-hosted recordings to Steve or the official service. Ask that instance’s operator for its privacy information.

2. Information processed

The recording feature captures audio rather than Discord video or screen sharing. It is not designed to archive general Discord text-channel conversations.

3. Purposes and recording controls

We process this information to record sessions requested or configured by server users, generate and deliver exports and transcripts, maintain settings, recover interrupted sessions, diagnose failures, protect the service, and carry out deletion. Optional audio/video sync saves a generated cue track, cue IDs, and timing markers. When you pair VTT Cameraman using a private read-only link, it receives session/cue IDs and timing to produce its flash and beep; that feed does not deliver voice recordings. A short sound announces recording starts by default, unless disabled or playback is unavailable; it does not collect consent. Automatic recording can be enabled through server settings, schedules, or event rules. The person enabling it must notify participants and obtain the required consent; the software does not collect an individual consent response from every speaker. If you do not want to be recorded, tell the organizer and avoid speaking in a channel while recording is active. Leaving stops future capture of your voice through that connection but does not erase audio already saved.

4. Official hosted storage and processing

Official recordings and generated files are stored on the operator’s RackNerd VPS. Transcription is optional and currently runs locally on that server using whisper.cpp and a configured model; it is not sent to a hosted transcription API by the current configuration. Transcripts may contain errors and sensitive details spoken in the session. A person with Manage Server permission can connect one Google Drive or Dropbox account to their Discord server. Each server uses only its own connection; servers never share an account. When a server sends an export to the cloud, or turns on automatic upload, the recording and its generated files are sent to that account and stored by that provider under its own policies. The person who connected the account controls those copies and can read them, even if they would not otherwise have had access to the recording. We store encrypted sign-in tokens for that account on the VPS. We use them only to write these files into a folder this app creates: The Witness cannot read, list, or change your other files. We do not use provider account data for advertising or sell or transfer it, and our staff do not read the tokens except as needed for security or legal reasons. Disconnecting with /cloud, or removing the bot from the server, deletes our tokens and asks the provider to revoke access. The Witness’s use of information received from Google APIs follows the Google API Services User Data Policy, including its Limited Use requirements. Self-hosted operators can change executables, models, hosting, and integrations, so these official-hosted details do not apply automatically to them.

5. Access and sharing

The hosted operator and people administering the host can access stored files for service operation and support. Discord carries bot commands, recording notices, and responses; exports delivered as Discord attachments are also stored by Discord. People receiving downloaded files or valid signed links can access the content. A signed link is not tied to a particular Discord account: anyone holding it can use the access it grants, including export or dashboard actions where available. Links normally expire after 24 hours, but new links can be issued while files remain stored. Keep links and downloaded archives private.

RackNerd provides the VPS, and Cloudflare fronts the public website. These providers process infrastructure and network data needed to host and deliver the service; Cloudflare also supplies network-error reporting headers, so browsers may send connection diagnostics to Cloudflare. We may disclose information when legally required or necessary to address security or abuse. The application has no advertising, data-sale, or model-training workflow. These pages load no third-party analytics or tracking resources and set no cookies. The current website configuration disables routine web access logs; error and application logs can still be produced. This is not a promise of anonymous use or end-to-end encryption of stored recordings.

6. Retention and deletion

Copies uploaded to a connected cloud account are outside these limits: they stay until the account holder deletes them, and disconnecting does not remove them. We cannot delete them on request, so ask the person who connected the account. Original recordings on the official hosted service are deleted automatically 31 days after they end, along with their local exports. This includes failed and interrupted recordings. Generated exports are cleaned after 48 hours unless active jobs or saved editor work require them; terminal export-job records expire after seven days. Unreferenced server-intro and custom start-sound versions are cleaned after 48 hours, and abandoned uploads after one hour. A custom start sound referenced by server settings is kept even when its mode is Default or Off. Do not interpret link expiry as a retention deadline or a guarantee of availability for the full period; download anything you want to keep. A server manager can set a shorter period for completed recordings, measured from completion. Each server also has a recording storage limit, and a server over it cannot start new recordings until space is freed. Cleanup runs periodically and skips active or busy sessions, so deletion may happen somewhat after the stated period.

A member with Manage Server permission can stop a recording and use /delete with its session ID and confirmation to remove the session directory, including its audio and local exports. This does not remove downloaded copies, Discord attachments, external uploads, server-wide intros, start sounds or settings, or all operational/job records. Contact the operator about removing server-wide audio and its saved settings. Separate recovery sessions and any independently retained copies need separate attention. Removing the bot from a server does not automatically delete prior recordings. We do not promise immediate erasure from every infrastructure copy or backup.

7. Requests and concerns

To request access, correction, deletion, or assistance with a recording, contact your server administrator or the hosted operator below. Include the server, approximate session date, and session ID if available, and explain your request without sending private audio publicly. We may need to verify your connection to the session and coordinate with its organizer to protect other participants’ information. Privacy rights depend on applicable law; nothing here limits rights that apply to you.

For the official hosted service, contact Steve (Straconis), the project owner, at support@thewitness.dev. For a self-hosted instance, contact its operator. Do not post recordings, transcripts, private download links, or other personal information publicly.

8. Policy changes

We will update this page and its effective date when the service or its data practices change. Self-hosted operators must explain their own changes to their users.