News & Updates
Announcements, product launches, and research from Cogerphere — including Cobentt, Openbentt, RRbench, and the published Meridian 0.1 paper.
Latest
New ResearchProject Kostant — a blueprint for fast, fully local LLMs on phone processors
New research snapshot from Cogerphere AI Labs. Kostant asks what it takes to build an Ollama/vLLM for phones: mobile-native 1–4B models, 4-bit weight kernels, phase-aware NPU/CPU/GPU execution, compressed KV-cache and speculative decoding. Resident quantized small models as the fast path.
Read more
Coming SoonCobentt Mobile Playground — coming soon (in testing)
Development is done — now in testing. Cobentt's mobile playground will bring documents, slides, sheets, deep research, chat, websites and smart contracts to your phone. Stay tuned on News and Instagram @cogerphere.
Read moreNewCobentt — documents, slides, sheets, deep research, chat, websites, and smart contracts in one workspace
Cobentt is now live at cobentt.vercel.app — an all-in-one AI workspace. Write documents, build slides and sheets, run deep research, chat, generate websites, and create smart contracts with AI. Mobile playground app coming soon — development complete, now in testing phase.
Read moreAll updates
ProductBeta Launch
May 25, 2026
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.
ProductNew
2026
Openbentt Phase 1 is here
A local-first AI workspace for LaTeX, PDFs, benchmarking, and fine-tuned small models — built for researchers, by researchers. Inference, history, and attachments stay on your machine.
ResearchUpcoming
Coming soon
RRbench (ResearchBench) — trust-first research verification
A unified platform for assessing credibility and reproducibility of scientific research — especially AI/ML papers. Multi-signal Trust Scores combine AI analysis, artifact integrity checks, expert review, and community validation.
ModelPaper
Published
Meridian 0.1 — LaTeX-aware academic proofreading SLM
Our published paper, Meridian 0.1: Why Data Design Beats Method Innovation, introduces Meridian-AC-Nano, a 751.6M-parameter multi-task academic assistant built for structured academic text and LaTeX-aware processing.
Research
Ongoing
Context Intelligence Framework
Adaptive compression and drift detection for LLM context windows — extending what models remember without losing coherence.