Shepherd Terms of Service

1. Overview

Shepherd helps people and communities put the right help within reach: a Discord bot and web support tracks for difficult moments, and a set of engines (such as Truth) available through the Shepherd site and, where offered, programmatic endpoints. These Terms govern your use of the Shepherd website, Discord bot, support flows, engines, credits, analytics, and related services.

2. Who may use Shepherd

Server owners and administrators are responsible for deciding whether Shepherd is appropriate for their community, configuring where it is active, and following Discord's rules and applicable law. Users may engage with Shepherd support prompts voluntarily.

3. Discord and server configuration

Shepherd operates through Discord and depends on permissions granted by server administrators. Admins control whether message scanning is enabled, which channels are configured, intervention behavior, and related settings. Removing Shepherd or changing its Discord permissions may disable some features.

4. Support boundaries

Shepherd is not therapy, medical care, legal advice, emergency response, or a substitute for trusted adults or qualified professionals. In an emergency or crisis, contact local emergency services, a crisis hotline, or a trusted person immediately.

5. Faith-based content

Shepherd may offer optional Christian faith-based encouragement, scripture, prayer prompts, and related next steps. Practical support remains available for users who do not want faith-based guidance.

6. Acceptable use

Do not use Shepherd to harass, shame, surveil, exploit, or target people. Do not attempt to bypass safety controls, extract private data, interfere with the service, or use Shepherd in ways that violate Discord's terms, community rules, or applicable law.

7. Credits and paid engine actions

Today nothing here costs credits. No credits are held, deducted, or charged, and there is no balance to run down.

What gates a run depends on which one it is, and “assessment” names two different things here. Submitting a claim for a real assessment, or asking Shepherd to read a source does model processing: it requires a signed-in approved account and is rate-limited to that account. The open developer endpoint is unauthenticated— anyone can call it, with no account and no sign-in — and is limited only by network address and input size. It never does model processing: it returns a fixed, deterministic stub response that labels itself as a stub, so it is not a paid engine action and nothing in this section applies to it. Shepherd’s own services also call the engines internally with a service credential; that is Shepherd calling itself, not access granted to anyone. This section describes the credit accounting Shepherd is building, so that the terms are published before the behaviour rather than after it. It takes effect only when that system is live, and the change will be dated here.

Once it is live, some engine actions (for example a Truth assessment or a source read) will consume credits. Credits measure model processing: 1,000 credits corresponds to US$1.00 of model cost, credits are whole numbers, and any nonzero cost rounds up so that work which cost something is never recorded as free.

Credits are currently issued by grant only. Shepherd does not sell credits today, and there is no checkout, price, or payment processor. A paid way to add credits is future work; these Terms will be updated before any purchasing opens, and the purchase, refund, tax, and billing terms that a paid service requires will be published at that time.

Credits belong to an individual signed-in account and are spent by that account. Anyone able to sign in to an account, or use an API credential issued for it, can spend its credits, so treat account access and API keys as you would a payment method.

8. When credits are consumed

This section describes the credit system once it is live; see section 7 for what applies today. Under it, credits are held when work starts and settled when it finishes:

• Held before work begins. A run reserves a high-side estimate up front. If the balance cannot cover it, the request is refused and nothing is charged. If the credits system itself cannot be reached, the request is refused rather than run unmetered.
• Settled once, at the end. A completed run is charged its actual measured cost, which may be less than the hold but never more: the hold is a ceiling, not an opening bid. A run that would pass it stops instead (see below), so no work can charge past what it reserved. Settlement happens only at a final state, and repeating the settlement never charges twice.
• Cached results cost nothing. A run served from an existing result settles at zero.
• Duplicates are not double-charged. Submitting the same work again while your own run is still in flight releases the second hold uncharged.
• Re-running previously completed work is charged again. Each run is its own paid unit, including a re-run of something assessed earlier.
• Work that fails partway can still cost money. If a run fails, times out, or is abandoned after it has already spent on model processing, that spend is charged and only the unspent remainder of the hold is released. A run that never spent anything releases in full.
• Retries accumulate. Where a run retries internally, the charge is the cumulative cost of those attempts, including any per-delivery third-party fee such as a transcript provider.
• Spending stops rather than overruns. A run that reaches the limit of what it reserved and funded is refused rather than allowed to spend past it.

When the credit system is live, your account page will show your balance, what is currently held, and the ledger of every hold and charge. That surface ships with it — these terms do not ask you to trust an accounting you cannot inspect.

9. What paid processing does and does not buy

Credits pay for processing, not for a particular answer. A Truth assessment that abstains, finds the available evidence insufficient, or returns a conclusion you disagree with has still done the work and still consumes what it consumed. Shepherd does not guarantee that a result will be favourable, complete, or suitable for any decision you make with it, and engine output is not professional, legal, medical, or pastoral advice.

10. Programmatic and API use

Where Shepherd exposes engine endpoints, you may call them with a credential issued to you, for your own use, within the published rate limits. You keep the rights you already hold in the content you submit, and you grant Shepherd permission to process it in order to run the action you asked for. You may use the output you receive. Do not use the API to build a competing derivative of the engines, to submit content you lack the rights or permission to submit, or to evade rate limits or the credits system. Access may be rate-limited, suspended, or withdrawn for abuse or non-payment, and endpoints may change as the engines change.

11. Changes and availability

Shepherd is evolving and may change, pause, or discontinue features at any time. We may update these Terms as the product changes. The updated date above shows when this page last changed.

12. Intellectual property and trademarks

™ 2026 SOVEREIGNify, LLC. All rights reserved. Shepherd, SOVEREIGNify, and related marks are trademarks of SOVEREIGNify, LLC. You may not copy, modify, or use Shepherd or SOVEREIGNify marks in a way that suggests endorsement, partnership, or ownership without written permission.

13. YouTube and third-party content

Shepherd surfaces music via YouTube API Services and embeds official videos through YouTube's player. By using Shepherd, you also agree to be bound by the YouTube Terms of Service. Videos remain the property of their owners and are shown through YouTube's official embedded player with attribution and a link to the source; Shepherd does not download, host, or modify any video or audio.

14. Contact

Questions about these Terms can be sent to hello@sovereignify.ai.