One composable API to detect grooming, sextortion, distress, deepfakes, TFGBV and more, across 13 harm categories across text, voice, image, and video. EU-resident. Zero retention. 32 languages. 1.1-1.5s typical end-to-end.
What Our Partners Say
Trusted by teams building safer products
"We don't want any more children to slip through the gaps and Tuteliq can help make that a reality."
Kerry Smith
CEO
"At SEEN we deliver personalised video at scale for some of the biggest brands in retail and financial services."
Ronald Griffin
"Instead of reacting to issues after they surface, we now have better visibility into potential risks before they escalate."
Ron Novak
CEO and Co-Founder
"The fast-evolving safeguarding environment led us to a dedicated tool."
Hugh Shepherd
Founder & CEO
"Tuteliq helps us keep our service away from underage users without requiring intrusive identity verification."
Eva Leach
"Tuteliq's AI-driven risk assessments let me identify vulnerabilities early, saving weeks of manual audits."
Gaurav Vashisth
Partner
"As an AI developer, one of my biggest concerns has always been how users interact with our agents, especially when young users are involved."
Dev Rishi Khare
Founder
FOR DEVELOPERS
How this differs from sentiment analysis and generic emotion AI
Sentiment APIs and emotion classifiers were built for marketing analytics. They score one message at a time and return polarity or a discrete emotion label. Neither output is a safeguarding decision. Behavioural detection scores the pattern, not the polarity.
Sentiment analysis API
Generic emotion AI
Tuteliq behavioural detection
WHAT IT SCORES Polarity (positive / neutral / negative) of a single message.
Basic emotion labels (joy, anger, sadness, fear) on a single message.
Behavioural signals, grooming tactics, coercion, distress, exploitation risk, across a conversation.
UNIT OF ANALYSIS One message at a time.
One message at a time.
Sequence-aware: role, turn order, conversation history, and age group.
SAFEGUARDING OUTPUT None. A negative score can mean a bad day or a crisis.
None. 'Sadness = 0.8' is not a safeguarding decision.
Risk level, tactic tags, evidence spans, recommended action.
AGE CALIBRATION No.
No.
Yes, age-group multipliers (under-13, 13-15, 16-17, 18+).
EVIDENCE RETURNED Score only.
Score + label.
Evidence spans, phase, confidence, recommended_action.
BEST FOR Product feedback, review triage, brand monitoring.
UX research, chatbot tone selection.
Child-safety, grooming / sextortion / distress detection, DSA / KOSA / OSA compliance.
How to choose
A quick decision guide for developers. If more than one row applies, pick the strongest safeguarding requirement. You can always layer sentiment on top of behavioural detection, but not the other way around.
You want polarity and topic on standalone text. AWS Comprehend, Google Natural Language, or Azure Text Analytics are fine here.
You want a coarse emotion label on a single utterance to adapt UI or agent tone. Any HuggingFace emotion classifier works.
You need grooming, sextortion, or distress detection with age calibration and evidence, polarity and single-message emotion cannot produce a safeguarding decision.
Route transcripts to the safety endpoint on every user turn and gate agent responses when vulnerability_level ≥ elevated. Sentiment on the last turn will miss the trajectory.
Regulators expect evidence, thresholds, and human review, not "sentiment = -0.7". Every response returns evidence spans, tactics, and recommended_action.
If there is no minor-facing surface, no crisis routing, and no regulator obligation, a generic API covers you. Revisit if you add minors, voice agents, or user-to-user chat.
Why teams choose Tuteliq over generic moderation APIs
Classifiers score words. Tuteliq scores behaviour, the trajectory of a conversation, the escalation of coercion, the drift of a user in distress.
Composable, not monolithic
Combine grooming, sextortion, distress, synthetic content and the other detectors across 13 harm categories into one call, pay only for what you use.
Minor-safeguarding by default
Age-calibrated thresholds, KOSA / DSA / OSA / eSafety alignment out of the box. Not a general-purpose 'toxicity' bucket.
Behavioural, not just artefact
Detects intent trajectories, grooming sequences, coercion loops, distress spirals, where classifiers see isolated messages.
Typical latency 1.1-1.5s
Fast enough to gate user-facing actions. No async polling, no separate 'moderation lake' to build.
32 languages, EU-resident
Culture-calibrated across 32 languages. Data processed in the EU only (Paris + EU Vertex AI). Zero retention.
Auditable by design
Every decision returns evidence, categories, and reasoning. DSA Statement-of-Reasons ready. SOC 2 Type II attestation and ISO 27001 certification in progress.
One API. Every category.
Call detectors individually, or compose them in a single request.
Building a chat app? See our sentiment analysis API for child-safety apps, pre-vulnerability detection with age calibration and evidence spans.
Frequently asked questions
What is a content moderation API?
A content moderation API is an HTTP endpoint that classifies user-generated content (text, images, audio, video) for policy violations, safety risks, or regulatory concerns. Tuteliq's moderation API goes further than generic classifiers by detecting behavioural sequences, grooming trajectories, coercion loops, sextortion escalations, not just isolated harmful words.
How is Tuteliq different from OpenAI Moderation or Perspective?
OpenAI Moderation and Perspective return category scores on single messages. Tuteliq is built for minor safeguarding: it reasons across conversation history, applies age-calibrated thresholds, returns human-auditable evidence, and covers behavioural harms (grooming sequences, coercive control, distress spirals) the others do not surface.
Do you store the content I send?
No. Tuteliq is zero-retention by default. Requests are processed in the EU (Paris via Mistral, EU Vertex AI for speech-to-text) and dropped after the response is returned. Only anonymised metadata is retained for billing and audit.
Which languages and regions are supported?
32 languages with cultural calibration, including English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Swedish, Turkish, Arabic, Japanese, Korean, and Mandarin. Compliance alignment covers KOSA (US), DSA (EU), Online Safety Act (UK), and Australia's eSafety codes.
How is pricing calculated?
Credit-based. Simple detectors (bullying, unsafe) cost 1 credit; behavioural detectors (grooming, sextortion) cost 3–5 credits; multi-modality composable calls charge per active detector. A 14-day free trial is available on any paid plan. See /pricing for details.
Is Tuteliq's moderation API suitable for adult platforms?
Yes. While the calibration is optimised for platforms that serve minors, adult-facing platforms use Tuteliq for TFGBV, sextortion, romance scams, non-consensual intimate imagery, and synthetic content detection, categories generic moderation APIs handle poorly.
Ship safer software this week.
14-day free trial on every plan. EU-resident. Zero retention.