Moderation workflows

When a submitted image prompt is flagged by the moderation classifier, this workflow holds the request, logs it, and posts the offending prompt to a Slack review channel…

Transcribes an inbound demo-request voicemail with ElevenLabs, extracts intent and contact details, creates or updates a HubSpot contact and deal.

An autonomous agent picks up each reported phishing email, detonates links in a sandbox, enriches indicators via web research, decides quarantine and escalation actions.

When a draft reply to an inbound social mention is generated, it is scored for toxicity by a HuggingFace classifier.

Inbound social DMs are toxicity-scored on arrival; hostile messages skip the bot and open a Zendesk ticket, while normal DMs get a screened AI reply auto-sent.

On a GitLab release, drafts a founder thread, generates a matching announcement image, and posts both to a Discord moderator channel for approval before publishing to social.

Ingests spam-complaint (FBL) webhooks from your ESP, tallies complaints per campaign in a rolling window.

On each inbound spam-complaint webhook it tallies complaints per campaign and domain, and once a campaign crosses the complaint-rate limit it quarantines the affected segment.

Ingests inbound comments and replies from your connected social platforms, classifies each for legal or PR risk.

After Replicate generates variants, an AI vision pass screens each one for brand and content issues.

When Dependabot opens a security PR, it pulls the linked advisory severity and, for critical or high CVEs.

An agent reviews a flagged offender's complete record against your moderation policy, drafts a ban recommendation with reasoning.

When a sanctioned member submits an appeal form, an agent pulls their full case history from Notion, weighs the appeal against the original offense.

Every moderation action a mod takes is captured, summarized by an agent into a clean case note.

Captures ban-appeal form submissions, pulls the appellant's original offense history from Postgres, has an LLM assess whether the appeal addresses the actual violations.

When a user's strike count crosses a hard threshold, this opens a Linear issue with the full offense history and pings the moderation lead in Discord so a human decision is…

When a caller hangs up or asks for a callback in the voice IVR, ElevenLabs speaks a callback offer and the call's intent is summarized and logged to the support queue so an agent…

Polls Honeycomb on a short interval for trace error-rate spikes, summarizes the affected traces with OpenAI, posts a digest to Slack.

Accepts inbound brand-misuse tips via webhook, scrapes and verifies the reported page, and routes confirmed infringements to legal in Airtable while auto-closing false reports.

Intercepts inbound image-generation requests over a webhook, classifies each prompt against your content policy.

An agent investigates a reported secret leak end to end — confirms the credential is live, traces what it unlocks across systems, drafts a rotation plan.

Scores incoming Intercom return requests against the customer's refund history in Stripe and Postgres.

An agent triages high-stakes mentions, drafts a brand-aligned reply, and runs it through HuggingFace sentiment and toxicity checks plus a human Slack sign-off before any…

Scheduled scan of new post comments classifies each by sentiment, drafts a matching reply, blocks any toxic draft, and auto-publishes the rest while logging everything to Notion.

When a new Replicate variant lands, it runs an OpenAI vision check against brand-safety rules and routes pass/fail: clean variants go to the review queue.

Before rendering, an LLM checks the brief copy against brand and policy rules; clean briefs render and archive to S3, flagged briefs route to a review queue instead.

Watches the abuse@ mailbox for user-reported phishing, opens every embedded link in an isolated Browserbase session to capture redirects and screenshots.

Screens submitted image prompts using a HuggingFace text-classification model instead of a third-party moderation API.

When Stripe flags a charge as elevated-risk, checks whether the same customer or payment method has open or recent refund requests in Postgres and Intercom.

Runs on a schedule to roll up the previous day's prompt-moderation log from Postgres, computes block rates and top violation categories, and delivers a digest to your team.

When a community member reacts to a Discord message with a designated emoji, instantly narrates that message with ElevenLabs, generates an audiogram.

On a webhook-event trigger, watches for abnormal posting through your Discord webhooks (spikes, off-hours bursts.

On each blocked prompt, this workflow counts a user's recent violations in Postgres and, once they cross a strike threshold, suspends their generation access and alerts the team.

Weekly, clusters refund requests in Postgres by shared payment fingerprint, address, and device to surface coordinated fraud rings.

Watches a Discord server for a sudden burst of negative messages in one channel and pages the comms owner in Slack with a context pack of the offending thread and tone summary.

Mines your competitors' newest 1-2 star App Store reviews, summarizes the complaints they share, and posts a short Slack digest highlighting gaps your product could win on.

Classifies each flagged Discord message for rule violations with an LLM and records the offense against the author in a Postgres offender ledger, returning a running strike count.

Processes a submitted verification form, checks the answers against an approval policy, and either grants the verified Discord role or flags the applicant for human review.

When a moderator adds a checkmark reaction to a helpful answer in a thread, the bot captures that question-and-answer pair and writes it into the Postgres FAQ so the bot can…

Watches your #introductions channel, uses AI to confirm a new member's intro is genuine and complete, then auto-assigns the verified Member role and welcomes them.

When a moderator runs a slash command in Discord, instantly summarizes the chosen channel over a chosen time window into a themed brief.

When a moderator adds a designated emoji reaction to any Discord message, an LLM converts that message into a structured Linear issue and threads the link back.

When a member files a /report or reacts with a flag, an agent reads the offending message in context, classifies the violation against your rules.

Runs each morning to query the offender ledger for users whose violations are accelerating, summarizes the pattern with an LLM, and posts a ranked digest to the moderator channel.

Each week it collects the questions the bot escalated or couldn't ground, clusters the recurring ones.

On a weekly schedule, an agent reads the past week's case log from Notion, summarizes moderation trends and repeat offenders.

An agent-driven concierge that answers new members' questions in a #start-here channel using your knowledge base, escalates anything it can't confidently answer to moderators.

When a Front conversation is closed, this extracts the issue, resolution, and sentiment.

Pulls fresh brand mentions across social platforms every hour, scores each for sentiment and urgency, drafts a tailored reply.

When a Zendesk ticket moves to solved, generate a personalized one-question CSAT email with OpenAI and send it to the customer, skipping internal and spam tickets.

Watches a shared support inbox, and for every inbound email creates a Front conversation with an AI-drafted reply attached.

On a schedule, queries your spam-complaint table for the last hour, flags any sending domain over the 0.1% complaint threshold.

When a warned member submits an appeal via webhook form, the agent pulls the original case from Postgres, assesses the appeal against it.

Cross-references each refund request against a Postgres ledger of prior refunds and Stripe charge data to flag serial refunders and duplicate claims before an agent issues money.

When the FAQ bot can't confidently answer a member's question from the docs, it pings a moderator in Discord and opens a Linear issue so the gap is tracked and resolved.

On a schedule, an agent sweeps the pending moderation queue, drafts a decision and rationale for each open case.

Lets moderators define or edit gated onboarding FAQ steps via a Discord slash command, validates the new step with OpenAI.

Before queued replies go out, each is toxicity-scored by HuggingFace; failing items are pulled from the queue and flagged in Airtable while passing items publish on schedule.

When Cloudflare reports a sudden bot-traffic spike, this checks GitLab audit events for token activity from the same source IPs in the same window and pages PagerDuty if…

Runs on a schedule, queries Datadog CI Visibility for each test's flaky rate over the last 7 days.