altered brilliance
B2C
TGX1-PRO
Player-facing product experience
Loading page
Loading page content
ALTERED BRILLIANCE IS LIVE.ALTERED BRILLIANCE IS LIVE ON GOOGLE PLAY.Read the product story →
Codename-TGX1 interprets the structured cognitive context created by tGiX, using a specialised gaming SLM rather than asking a general-purpose model to decode raw telemetry.
The current Android release sends authenticated Game Intelligence requests through a Supabase Edge Function. That server-side boundary validates the conversation and model choice, keeps the provider key out of the client, and routes the request through OpenRouter.
Google Gemini 2.5 Flash is the current default proxy model, with GPT-4o mini allowed by the service. This mirrors the future TGX1 input pipeline while proprietary TGX1 training and on-device deployment continue.
Every TGX1 variant interprets structured context from tGiX. The model code signals the product environment and the people it is built to serve.
The player-facing TGX1 variant for Altered Brilliance, bringing tGiX-produced context into the product experience.
The development variant for studio workflows and the tGiX marketplace, bridging player-facing and game-development-facing data.
The enterprise variant of the TGX1 model family, scoped for organisations building on the structured context layer.
The education variant of the TGX1 model family, designed for learning-focused programmes and partner environments.
The competitive-performance variant of the TGX1 model family for eSports organisations and their workflows.
TGX1 is a unified SLM family, with model variants assigned to the commercial and product contexts where they create the most value.
altered brilliance
B2C
Player-facing product experience
tGiX marketplace
B2B2C
Player product and marketplace development for studios
Game Studios
B2B
Player-facing data and game-development-facing data
Enterprise
B2B
Enterprise product and platform environments
Education
B2B
Education programmes and partner environments
eSports
B2B
Competitive-performance environments
altered originals
B2C
Original player experiences and development workflows
TGX1 is being developed as specialized edge intelligence. The current app release validates the conversational product loop through an authenticated cloud proxy while on-device deployment remains a planned stage.
Authenticated app requests pass through a Supabase Edge Function to an allow-listed OpenRouter model, keeping provider credentials outside the mobile client.
The product assembles approved context from the Gamer Index, progression, library, sessions, and focus history before requesting an interpretation.
The proprietary 3.8B TGX1 foundation has completed its first training level. Specialized variants, benchmarking, and on-device execution remain subsequent stages.
| Specification Parameter | Value / Standard | Technical Details |
|---|---|---|
| Proprietary foundation | 3.8B / 4-bit target | Phi-4-mini-based TGX1 foundation with Level 1 training completed. |
| Current app runtime | Authenticated cloud proxy | Supabase Edge Function routing to allow-listed OpenRouter models. |
| Current default proxy | Gemini 2.5 Flash | GPT-4o mini is also allowed by the secured service. |
| Factored context | Index · Progress · Games · Sessions · Focus | Player-created context assembled before interpretation. |
| On-device TGX1 | Planned | Porting, benchmarks, specialized variants, and edge execution remain future stages. |
| Model family direction | PRO · DEV · CORP · EDU · ESP | Focused variants aligned to player, studio, enterprise, education, and esports environments. |
Keep promotion closed until the model produces valid, evidence-bounded outputs under frozen evaluation.
Reviewed August 2026 · no fixed release dates
View portfolio roadmapPublicly shipped or part of the current product foundation.
Dataset, training, adapter, manifest, deterministic evaluation scripts, and raw-output foundations exist.
GamePilot uses authenticated allow-listed cloud model routing today; TGX-1 is not represented as its current production model.
Source exists, but runtime, packaged, or release acceptance is incomplete.
The latest base and adapter evaluation produced zero valid structured outputs across 30 frozen cases.
Dependency: No candidate advances until schema and evidence adherence pass the frozen evaluation.
Approved work is actively being built or hardened.
Fix structured-output adherence, dataset lineage, privacy, evaluation boundaries, and explicit release criteria before further promotion.
Prepare consent-first, minimized, reviewed gameplay evidence without treating raw collection as automatically trainable.
Intended direction, subject to validation, sequencing, and change.
Evaluate player, developer, enterprise, education, and esports variants only after the shared foundation qualifies.
Explore edge execution and native GamePilot or Majestic integration after quality, privacy, and benchmark gates pass.
Scope a model-family conversation around your product, studio, enterprise, education, or eSports environment.
Model role, evidence boundaries, API access, and product integration.