Safety by design.

Empowering developers and builders to ship AI that children can actually use.

The compliance and safety layer for generative AI built for children under 13. Independent of any model. Provable to any regulator.

See the safety stack run Request technical access

Not a developer? AngelQ is the real solution for AI-savvy parents who want to empower their kids with AI in safe and healthy ways — and everything on this page is the work you are welcome to check.

An independent orchestration layer between any LLM and the child.

Swap models, change rules, prove compliance, without rebuilding safety each time.

Any LLM

OpenAI · Gemini · Claude · Grok

AngelQ safety orchestration

PII redactInput classifyIntent gate Vision scoreContent scoreDeferral

Child

An age-calibrated experience

Auditability

Prove to any regulator exactly which safety rules governed any child interaction.

Trained on real child conversations

The safety classifiers are fine-tuned on conversations children actually had. Not synthetic, and not transferable from an adult product.

Safety runs alongside generation

The checks do not queue behind the model. Vision and content classifiers fire in parallel while the answer is being drafted.

Every child's message, from the moment it arrives to the moment they hear an answer.

This is the live system, captured from the running app rather than drawn for a slide. Every child's message takes this path.

Every child's message, node by node A message arrives. Personal details are hidden. Hurtful wording is softened. The system reads it as a child would, asks what is actually being requested, decides whether it is safe, and only then do the seven gates open.

Hundreds of checks, running in real time Every component in the conversational path fires as one message travels through it, fast enough that a child never notices it happened.

Image safety gets equally intense treatment Every picture is scored on safety, relevance, and quality before a child sees it, so an unintended exposure never reaches the screen.

439components in the conversational flow

540connections between them

7gates that must all open

See what a brand can build on this foundation on the Brands page. Technical access starts with a scoped review. Request access

KidRails, the foundation of our guardrails, has been open sourced.

Anyone can use it to make their own apps safer, and our own work is transparent to the world because of it. AngelQ and Arcee AI built it together and published the code, the model, and the dataset. The technical summary the two teams wrote is the place to start.

Read the technical summary Open the repository

Where it lives

Code
github.com/arcee-ai/KidRails

Model
huggingface.co/arcee-ai/KidRails

Dataset
KidRails-Dataset