Perspective

Score text for toxicity and related attributes with Google's Perspective API, and submit suggested scores as feedback. Built for moderation and community-health workflows.

2 actions available

New comment submitted to a moderated space

Agent scores the comment for toxicity with Analyze Comment

Agent clears comments below the toxicity threshold

Agent publishes the cleared comment

Moderators get a queue of the flagged comments

A borderline comment routes to a moderator to approve or remove

What This Integration Enables

Perspective gives a flow a toxicity signal. Analyze Comment scores text across attributes like toxicity, insult, and threat, and the feedback action submits suggested scores back to improve the models over time. It is narrow by design: a fast, consistent read on whether a piece of text is likely harmful. A consistent score is what lets moderation scale: comments that are clearly fine publish, comments that are clearly harmful are blocked, and the score draws the line the same way every time. The borderline band is where judgment lives, and that belongs to a person. An orchestration layer clears the clear cases and routes the borderline ones to a moderator, and FlowRunner is built for that layer.

Without FlowRunner

Every comment reviewed Moderators read all submissions to catch the bad ones

Inconsistent calls Different moderators apply different thresholds

Toxicity reaches the feed Harmful comments appear before anyone catches them

With FlowRunner

Automatic first pass Clear comments publish, only flags reach moderators

Consistent scoring A single toxicity signal applied to every comment

Caught before publish Flagged comments held for review before they appear

Use Case Scenarios

Comment Moderation

New comments arrive in a moderated space. The agent scores each one for toxicity, publishes the comments below the threshold, and holds the rest. Moderators review a queue of flagged comments instead of reading every submission. Clearly-fine comments appear immediately, and harmful ones are caught before they publish.

Community Health Monitoring

A community platform needs a read on whether discussion is turning toxic. The agent scores incoming content and tracks the trend, surfacing spikes to the community team. The team acts on a signal rather than discovering a problem after it has spread, and individual borderline posts still go to a moderator.

Pre-Publish Content Check

User-submitted text is checked before it goes live. The agent scores it and clears the safe content automatically. Anything in the borderline band is routed to a person with the score attached, so the model handles the obvious cases and a human owns the judgment calls where reasonable people might disagree.

Human-in-Loop Highlight

Toxicity scoring is consistent and still blunt in the middle of the range, which is exactly where moderation decisions are hardest. When Perspective scores a comment in the borderline band, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step rather than letting a threshold decide: the agent pauses, presents the comment and its scores, and sends it to a moderator via Slack. They approve or remove it. The clear cases are handled automatically; a person owns the calls where context matters.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via Slack

Human decides

Agent resumes with decision

Agent Capabilities

2 actions

Analysis

Analyze Comment Analyzes a piece of text for a set of attributes such as toxicity, insults, threats, and profanity using Google's Perspective Comment Analyzer API.

Feedback

Suggest Comment Score Submits a suggested score for a comment to provide feedback to the Perspective models (the SuggestCommentScore method).

Start building with Perspective

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us