Blog
Product launches, research philosophy, and long-form updates from the Cogerphere team.
Project Kostant: A Blueprint for Fast, Fully Local LLMs on Phone Processors
Cogerphere AI Labs releases a research thesis on what it takes to build an Ollama/vLLM for phones — 1–4B mobile-native models, 4-bit kernels, phase-aware NPU/CPU/GPU, compressed KV-cache and speculative decoding.
Read article
Meridian 0.1: Why Data Design Beats Method Innovation
Our paper on a LaTeX-aware academic proofreading SLM is published. We introduce Meridian-AC-Nano, a 751.6M-parameter multi-task academic assistant, and make the case that data design and task formulation can deliver large gains without bigger models.
Read article
RRbench (ResearchBench) — trust-first verification for scientific research
ResearchBench is Cogerphere’s upcoming platform for assessing credibility and reproducibility of scientific research — multi-signal Trust Scores backed by AI analysis, artifact integrity, expert review, and community validation.
Read article
Cobentt — one AI workspace for documents, slides, sheets, deep research, chat, websites, and smart contracts
Cobentt is now live at cobentt.vercel.app. Write documents, build slides and sheets, run deep research, chat, generate websites, and draft smart contracts — all in one workspace.
Read article
Openbentt Beta Launch: Building the Local-First AI Research Workspace We Wish Existed
Today marks the beta launch of Openbentt — a local-first AI research workspace built for researchers, developers, and privacy-conscious professionals who need more than a chat interface.
Read article