const res = await client.moderation.check({
entity_type: "listing",
entity_id: "listing-8231",
entity_creator_id: "user-412",
moderation_payload: {
texts: ["Pay me on cash app, 50% off"],
images: ["https://cdn.app.com/8231.jpg"],
},
config_key: "marketplace",
});
// res.recommended_action -> "remove"AI content moderation for trust & safety teams.
Detect 35+ harm categories across text, images, video, audio, and livestreams in 50+ languages, at 98% accuracy. Nora, your moderation agent, reviews every flag in context and recommends or takes action based on your policies.
Review queueLive
High-performance detection at scale.
Read Moderation benchmark- 1B+end users across products
- 99.999%uptime SLA
- Globaledge network
- 8B+unique moderation requests/month
- 2B+images moderated/month
- 120M+policy violations identified/month
- 35+ harm categories supported
- 98% harmful content catch rate
- sub-50ms text moderation latency
- 300ms to score a live video frame
Request distribution
Share of requests by modality
- Text45%
- Images30%
- Usernames15%
- Video10%
Most tools can't keep up with attacks.
With Stream, your team writes the fix.
Without Stream
A new attack appears.
Scammers change wording, links, or images to slip past your filters.
Users report it before your tools catch it.
The damage is already done.
You wait on a vendor or a model retrain.
Weeks pass before the fix ships.
Attackers have already moved on.
The cycle starts again.
With Stream
New spam shows up.
A link-in-bio adult promo hits your comments. Nora flags the problem.
Write the policy.
Describe the harm in plain language and Stream turns it into a label.
Test it.
Run the policy on past content and see what it would have caught.
Activate it.
Spam drops back to baseline the same day.
From new spam to new rule in minutes.
Write the policy in plain language, test it on past content, and activate it.
Visual rule builder
Combine AI labels, user history, and thresholds into if/then rules without code.
Scam responseEnabledWHENLabel : ScamANDPrior warnings ≥ 2THENBan userCustom actions
Remove, warn, mute, ban, or send to your team, per policy.
Harassment policyHigh severityActions for · Harassment · HighRemoveBounceBanSend to teamSaved to audit logReversible by moderatorsFull audit log
Every decision, with who or what made it and why.
14:02Removed #48213Nora13:57Policy "Adult promo" activatedA. Rossi13:41Approved ban, user_412J. Kim13:20Kept #47990M. Diaz
Meet Nora, your moderation agent.
Older tools send every flag to a human, and the queue keeps growing. Nora reviews flags the way your best moderator would, resolving 80% of items.
For every flag, Nora checks:
Context
Does the flag hold up in the context of the conversation?
History
Has this user done this before?
Prior actions
What corrective action has already been taken?
Policy
Applies your policy skills, then acts or recommends.
Nora decisionFlag #48213 · Harassment · classifier 0.81
“you're actually so bad at this, uninstall before you embarrass yourself again”
ContextDirected at one player after a lost match. Not banter between friends.
History3 harassment flags in 14 days. 1 upheld.
Prior actionsWarned on Sep 30.
Policy skillPlayer conduct v4 · repeat targeted insults
Action takenMessage removed · 24h mute
Nora decisionFlag #48213 · Harassment · classifier 0.81
“you're actually so bad at this, uninstall before you embarrass yourself again”
ContextDirected at one player after a lost match. Not banter between friends.
History3 harassment flags in 14 days. 1 upheld.
Prior actionsWarned on Sep 30.
Policy skillPlayer conduct v4 · repeat targeted insults
RecommendedRemove message · 24h muteApproveOverride
Moderate any content, from any platform.
Use it on its own, or with Stream Chat, Video, and Feeds.
- TextScams, spam, harassment, and PII in 50+ languages.
- ImageExplicit, violent, and unsafe images, plus the text inside them.
- VideoUploaded and recorded video, sampled every second.
- AudioVoice messages and audio files, checked with your text policy.
- LivestreamsLive video, calls, and audio rooms, checked while they happen.
Compliance-ready by design.
Audit logs and decision reasoning support transparency reporting.
Digital Services Act
Protect users and reduce legal risk under Europe's strictest digital rules.
EUOnline Safety Act
Meet new safety standards and enforce age-appropriate protections.
UK & AUCOPPA & CSAM
Flag content automatically to protect minors and prevent child exploitation.
USEU AI Act
Show transparency, human oversight, and risk controls for AI decisions.
EU
Stream moderation FAQs.
Text, images, video, audio, and livestreams, plus what your own AI generates. You send content from any app or backend through one API, and every content type follows the same policy.
Nora reviews each flagged item. She checks whether the flag holds up in context, whether the user has done it before and what action was already taken, then applies your policy. You decide which policies she can act on alone.
Yes. Set a policy to Approve first and Nora prepares each decision with her reasoning for a moderator to accept or change. Switch it to Auto-act when you trust the results.
In minutes. Write the new rule in plain language, test it on past content, and activate it. No model retraining and no vendor ticket.
AI checks every item in milliseconds and in 50+ languages, which no human team can do. People are still best at judgment calls and appeals. Stream combines both: AI detects, Nora reviews, and your team handles what she escalates.
Classic AI moderation uses classifiers trained on fixed harm categories, which are fast and cheap at high volume. LLM-based moderation reads intent and context and follows labels you write in plain language. Stream runs both under one policy.
Pricing is based on usage across text, images, video, and audio. See plans and volume pricing on the pricing page.
Test on your own content.
Run your last 1,000 flagged items through Stream's API at no cost.