> For the complete documentation index, see [llms.txt](https://eirin-ai.gitbook.io/eirin-ai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://eirin-ai.gitbook.io/eirin-ai-docs/trust-and-safety/guardrails-and-permissions.md). # Guardrails and permissions The safety guardrails Eirin runs with, what it will never do without your approval, and exactly what store access allows. Eirin is built so that nothing reaches your players without your say-so. These guardrails are on from day one, and you can see and edit them on the **Playbook** tab of every agent. ## Default guardrails | Rule | What it means | | ------------------------ | ------------------------------------------------------------------------------------------------------------------------------ | | **Read-only by default** | A new agent reads reviews and drafts replies. It posts nothing until you connect a store and approve, or switch auto-reply on. | | **One manager only** | Only the manager can give instructions. Other senders are ignored. | | **No surprise posts** | No public responses without your explicit approval, unless you turned auto-reply on for that store. | | **PII protection** | Never exposes customer data in replies. | | **Transparent logging** | Every action is logged and auditable. | ## The reply playbook Every agent starts with eight reply guardrails. They are enforced on every reply, whether it posts on its own or waits for you: 1. Never publicly respond to users without human approval *(default, cannot be removed)* 2. Do not share unreleased roadmap items *(default, cannot be removed)* 3. Never promise fixes, dates, or compensation 4. Never argue with or dismiss criticism, acknowledge it 5. Never invent an answer, say you'll check and escalate 6. Never delete criticism of the game (only ToS violations) 7. Disclose being an AI when asked directly 8. Max one reply to trolls or bait, then disengage Rules 3 to 8 can be removed, and you can add your own. A second list, **guardrails for how Eirin generates reports**, holds topics to leave out of reports entirely, for example "leave out anything about monetisation or pay-to-win mechanics". ![The Playbook tab](https://592420414-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGQPBrXn3FlQhSCpExATE%2Fuploads%2Fgit-blob-c1bbff3334558c3e859bf08920a3a0b753797a01%2F28_Profile_playbook.png?alt=media) Each reply in the queue shows which rules it followed, so you can see the playbook at work. See [Approve and post replies](/eirin-ai-docs/working-with-eirin/approve-and-post-replies.md). ## What Eirin will never do (without your approval) {% hint style="success" %} **Read-only guarantee.** Without your approval, Eirin will never: * Post to any platform * Change its own instructions or permissions * Contact anyone other than you * Take any action not approved by you * Access data beyond what you've configured {% endhint %} ## How replies stay under your control * **Off by default:** auto-reply on the App Store and Play Store is off until you turn it on for that store, and you can switch it off any time under **Reply setup → Store connections**. * **Judgement calls always wait:** even with auto-reply on, replies that need judgement (crash reports, monetisation complaints, roadmap questions) wait for you under **Needs you**. * **Your rules:** replies follow your saved reply rules and playbook guardrails. * **Preview first:** generate previews on real reviews and test any review before anything goes live. * **Pause at any time:** **Disable** an agent from its profile and it stops drafting and posting immediately. Its settings are kept. ## Who can control Eirin Only the manager can configure agents and give instructions. Team members can read the queue and receive forwarded reports. ## Store access {% tabs %} {% tab title="Apple App Store" %} * **Least privilege:** the Customer Support role only reads and replies to reviews. * We cannot change your app, pricing, builds, or finances. * Your .p8 is uploaded over an encrypted connection and stored encrypted at rest. You never email it. * **Revoke any time:** Users and Access → Integrations → your key → Revoke. Setup guide: [Apple App Store](/eirin-ai-docs/connect-your-sources/apple-app-store.md) {% endtab %} {% tab title="Google Play" %} * **Read-only review data:** we read review text, ratings, and reply metadata. * **Reply to reviews:** we can post replies to reviews on your behalf. * We cannot change your app, pricing, builds, finances, or store listing. * **Revoke any time:** Users and permissions → our service account → Remove access. Setup guide: [Google Play](/eirin-ai-docs/connect-your-sources/google-play.md) {% endtab %} {% tab title="Discord" %} * Eirin only reads the channels you choose with `/setup-channels`. * Bot messages are captured but filtered out before analysis. * Replies are off until you switch them on under **Reply setup → Social connections**; the bot then posts in-channel. Setup guide: [Discord](/eirin-ai-docs/connect-your-sources/discord.md) {% endtab %} {% tab title="Instagram" %} * You authorise your business or creator account through Instagram's own login. Eirin never sees your password. * Permissions cover reading your posts and comments, and managing comments so Eirin can reply. Eirin does not publish posts or stories. * Replies are off until you switch them on under **Reply setup → Social connections**. Replies never contain links and go to new comments only. * **Revoke any time:** **Disconnect / use a different account** on the Sources tab, or remove Lysto under Instagram's **Apps and websites**. Setup guide: [Instagram](/eirin-ai-docs/connect-your-sources/instagram.md) {% endtab %} {% tab title="Support email" %} * You connect the support mailbox through Google's own sign-in. Eirin never sees the password. * Eirin reads the mailbox and replies in existing threads. It does not label, archive or delete mail, and never starts a new thread. * **Every reply is approved by a person.** Support mail is private correspondence, so there is no automatic sending, whatever the agent's other settings say. Eirin also never skips a support email on its own. * **Revoke any time:** **Disconnect / use a different mailbox** on the Sources tab stops sending immediately, or remove Lysto under the Google account's **Third-party apps and services**. Setup guide: [Support email](/eirin-ai-docs/connect-your-sources/support-email.md) {% endtab %} {% endtabs %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://eirin-ai.gitbook.io/eirin-ai-docs/trust-and-safety/guardrails-and-permissions.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.