RogerAI is organized as a model family, not a single generic assistant. Each model has a specific role: deep reasoning, daily assistance, engineering work, semantic indexing, retrieval ranking, and vision-based document understanding.
The catalog below explains what each model is designed to do, the kind of work it should handle, and how it contributes to the RogerAI product ecosystem.
Flagship reasoning and language model. Apex is the quality-first model for high-value enterprise work where context, precision, and structured thinking matter more than speed.
Balanced daily assistant model. Core is the default operating model for everyday business interaction in Arabic and English.
Engineering intelligence model. Forge supports technical execution, repository work, automation, and operational debugging.
Semantic indexing model. Index turns text into searchable meaning so RogerAI can retrieve the right knowledge instead of only matching keywords.
Retrieval quality model. Rank improves search quality by reranking candidate passages before the answer is generated.
Vision and document understanding model. Scan connects images, screenshots, forms, and scanned pages to structured AI workflows.
RogerAI routes work to the model layer that fits the task instead of forcing every request into one model.
Apex and Core handle language, analysis, drafting, decisions, and daily assistant workflows.
Forge supports code, automation, logs, tests, repository analysis, and DevOps tasks.
Index and Rank power semantic retrieval, relevance scoring, and grounded answers.
Scan reads screenshots, images, scanned pages, tables, forms, and visual documents.
These requirements are shown after the model catalog because they support deployment. They are not the model explanation.
Modern Linux server environment.
Python 3.10 or 3.11 with torch, transformers, tokenizers, safetensors, and accelerate.
NVMe SSD storage with enough capacity for model files, cache, logs, and future updates.
Persistent inference service such as vLLM, TGI, or an equivalent private serving layer.