Classification that survives adversaries.
AEGIS is the classification engine behind FrameBright. Built for content that was designed to get past filters. Licensable as a network-edge, device-embedded, or enterprise component.
FrameBright.ai is the technology, licensing, and partner surface for FrameBright. Not a parental control app. The classification infrastructure that runs underneath our consumer brand, our community directory, and a growing list of platform partners.
Nine detectors. One verdict.
AEGIS runs nine parallel detectors against every inbound piece of content. Each detector is purpose-built for its signal channel. Results compose into a single classification verdict with confidence, evidence, and a chain of signatures. Click any detector to see what it contributes on a real sample.
Classification that works on content designed to fool it.
Most content classification is benchmarked on clean datasets. AEGIS was built against content engineered to evade filters. Watch a side-by-side on three adversarial patterns: encoded media, cross-modal evasion, and context-collapse attacks.
Four deployment surfaces. One engine.
AEGIS runs where the content does. At the CPE for ISPs. At the hardware layer for device OEMs. At the router for network vendors. As a managed API for enterprise customers. Pick a surface to see the architecture, the integration effort, and the performance shape.
Already running on real content, at real scale.
Before selling AEGIS to platforms, we deployed it on our own properties. FrameBright.com curates a kids-and-parents content library. ParentProof runs a community content-classification directory. The FrameBright mobile app filters content in real time. Three live-in-the-wild deployments, one classification engine.
Games that re-theme themselves to the kid playing them.
AEGIS classifies what content is. The generative gaming layer changes what content becomes. Same learning objective, same difficulty curve, restructured into the theme, vocabulary, and characters a specific kid actually cares about. Powered by R1 as the agent runtime, governed by the same trust stack as the rest of FrameBright.
Same lesson, the kid's world.
A spelling game becomes a dinosaur dig, an ocean dive, or a wizard's spell. Vocabulary, character names, settings, and reward art are regenerated per session from the kid's stated and observed interests.
The mechanic bends to the player.
A kid who reads at level 1.4 gets shorter prompts and bigger reward loops. A kid who skips intros gets a cold open. The skeleton of the game stays; the pacing, prompt length, and reward cadence flex per player.
Generated, then verified.
Every generated frame, line of dialogue, and reward asset passes AEGIS before it reaches the kid. Generation without classification is the failure mode of consumer LLMs in kid contexts. We close that loop.
A 6 by 6 board, figurines that know who they are, games that rewrite themselves. In progress · Not ready for sale
A physical board with thirty-six squares, each holding a magnet, an NFC reader, and an RGBW light. Figurines have unique IDs. The board talks to the FrameBright app and the FrameBright Smart TV app to host generative games where the screen reacts to what is on the table and the table reacts to what is on the screen.
Notify me when pre-orders open →
Hardware development is underway. No funds are taken at this time. Sign up to be notified when pre-orders open.
A team that has classified adversarial content at scale before.
FrameBright was founded by people with direct experience running content classification at the hardest edge of consumer content. That experience informs AEGIS's design decisions; this is not classification theory.
Four licensing paths. Pick the one that fits.
AEGIS is licensed, not sold by the seat. Four commercial paths match the four deployment modes. Pick a path to see the commercial shape, typical deal structure, and what integration looks like.
Five things most content classification products get wrong.
Not features. Specific architectural decisions that make AEGIS different from legacy DNS blocklists, upstream CDN filtering, LLM-based moderation, and commodity classification APIs.
Purpose-built detectors, not an LLM wrapper.
Most modern content moderation tools are thin shells over a general-purpose LLM. AEGIS uses nine purpose-built detectors, each tuned for a specific signal channel (visual, audio, text, metadata, cross-modal, semantic, contextual, linguistic, behavioral). LLM-based moderation is one input to AEGIS, not its substitute. Purpose-built detectors catch patterns LLMs miss and cost a fraction of the inference.
Built against adversarial content, not clean datasets.
The hardest problem in content classification is content designed to fool the classifier. AEGIS was developed against content engineered to evade detection, not benchmarked on clean academic datasets. The founder's background at MindGeek running classification on Pornhub at 45M+ MAU shaped how AEGIS handles signal-inversion, encoding tricks, cross-modal evasion, and context collapse.
Classification at the distribution edge, not the application layer.
Legacy content filtering happens at the application layer, after content has already been delivered to the device. AEGIS is designed to run at the network edge (CPE, router, device hardware) so classification happens before the content reaches the application. This is a materially different deployment posture, and it is the reason AEGIS can be embedded by ISPs and OEMs in ways application-layer products cannot.
Licensable, not rentable.
AEGIS is licensed as a component partners embed in their own products, not rented as a third-party API they call. Partners own their integration, own their latency, own their customer relationship. This is closer to how classification infrastructure has historically been procured by major telcos and hardware OEMs than to the modern SaaS moderation API model.
Proven in the wild before sold to partners.
AEGIS did not launch as a B2B API with a pitch deck. It launched running on FrameBright's own consumer properties (the .com site, the mobile app, ParentProof), handling real content at real scale, with real parents and real kids depending on the classification being right. Partner conversations start from a product that has already been stress-tested on the hardest constituency in the market.
FrameBright.ai connects to a broader trust stack. In development
FrameBright.ai stands alone as a platform-licensing surface. Optional integrations are available when partner deployments need them.
Status (Q3 2026 roadmap): The R1 reasoning pass-through, RelayOne partner-governance routing, TrueCom royalty receipts, Heroa sovereign runtime, and ParentProof live-in-the-wild AEGIS calls described below are on the integration roadmap. AEGIS classification ships standalone today; cross-product wiring lands progressively as partner deployments require it. Enterprise licensing →
Agent framework for reasoning passes
R1 is the agent framework AEGIS calls when classification requires reasoning steps (e.g., cross-modal content analysis, contextual understanding, edge cases). AEGIS uses R1 as a targeted tool, not a general-purpose answer engine.
Governance for partner deployments
Enterprise partners deploying AEGIS under their own governance posture use RelayOne as the control plane. RelayOne enforces sovereignty, tenant separation, and audit evidence for the classification calls.
Signed receipts for royalty audit
Licensing transactions and partner royalty settlement run through TrueCom. Every classified content event emits a signed TrueCom-compatible receipt that partners can audit against invoices.
Sovereign runtime option
For partners deploying AEGIS in their own cloud footprint, Heroa is the optional managed runtime. Sovereign BC-Canadian deployment is available for Canadian telecom and public-sector partners via the Heroa sovereign substrate.
Live-in-the-wild proof
The consumer brand (FrameBright.com) and the community directory (ParentProof) are the live-in-the-wild deployments of the classification engine. Partners evaluating AEGIS can see it working on real content, at real scale, before they sign.
Works without the rest.
Partners who only want AEGIS get AEGIS. The connected stack is available to partners whose deployment posture benefits from it, never a precondition for licensing.
Methodology, not marketing.
Partners and technical evaluators deserve more than a data sheet. Our research page has methodology papers, benchmark disclosures, and ongoing work.
Two paths. Pick one.
Licensing conversation if you are evaluating AEGIS for a platform deployment. Partnership exploration if the fit is less defined.
Book a licensing conversation
30 minutes. We walk through AEGIS, your integration surface, and a rough commercial shape. No spec doc required up front.
Open a partnership exploration
Less structured. Open-ended conversation for researchers, press, and partners whose use case is outside the standard four licensing paths.