Private AI chat with verifiable citations, format-preserving translation of Word, Excel, PowerPoint and DXF documents, PDF-to-Word conversion, adversarial contract audits, and a grounded knowledge base — one unified, provider-agnostic agent runtime.
Under the hood
The Corporate AI Portal is our system for technical documents, data analysis and engineering insight — delivered as consulting, auditing, and managed services over a self-hosted AI workspace. Six modules share one governed, provider-swappable agent runtime.
Streaming AI chat with clickable citations, multimodal context, conversational memory and full session export. An agent that asks clarifying questions instead of guessing.
Word, PowerPoint, Excel and Markdown — translated and repacked with the original layout, fonts and structure preserved. Even DXF (CAD) drawings are re-encoded with structure retained. PDF is converted to Word, because a PDF stores positioned glyphs rather than paragraph structure. Terminology enforced by a multi-agent glossary pipeline.
A four-stage adversarial pipeline — Router, Consistency Auditor, an adversarial Challenger and an Executive Synthesizer — auditing seven dimensions, delivered as a consulting-grade audit report with a verified conflict matrix and redline recommendations.
Grounded Q&A over company documents with inline citation pills and a structured document viewer. Built on the vectorless RAG PageIndex technology — reasoning over the document's hierarchical structure instead of vector chunks. Self-hosted — company data stays on your side.
A full modern agentic stack: plan mode with approval cards, multi-agent teams, slash commands, a rich tool ecosystem, web-search cascade with citations — aligned with the latest agentic AI practices, protected by token budgets and approval gates.
Bring your own models from Google Vertex AI or any OpenAI-compatible provider; per-user limits, token and cost telemetry, and an append-only audit log of every command.
Every feature runs through a single agentic harness built on Microsoft.Extensions.AI: provider-swappable LLM engines, middleware-grade resilience — exponential backoff, circuit breakers, per-run token budgets — deterministic document codecs, loop detection, and human-in-the-loop approvals. Deployable as one container with a self-hosted RAG sidecar.
OpenSky develops open-source tools and utilities — free to use, fork and contribute to on GitHub.
Cross-platform desktop application for batch metadata and author management in Microsoft Office documents, built on Avalonia UI and .NET 8.
A lightweight, standalone desktop Markdown editor with WYSIWYG/Markup modes, LaTeX/Mermaid support, and seamless Word (DOCX) and Excel (XLSX) import/export integration.
Follow OpenSky on GitHub to keep updated.
Founder & CEO
Andrey Obushev is a Senior Executive and Nuclear Engineer with over 20 years of leadership experience managing large-scale projects in the international nuclear power industry — key contributions to the Russian ITER design team, and executive roles as Technical Director and Director for Licensing and Quality in the construction of the Akkuyu NPP in Turkey and the Paks II NPP in Hungary.
The Corporate AI Portal grew out of that background: AI that has to be governed, auditable and verifiable before it is trusted with real work.
A Bauman Moscow State Technical University alumnus with a lifelong interest in coding and analytics, Andrey founded OpenSky to carry the standards he worked to — nuclear safety, quality assurance, systems engineering — into enterprise AI.
Engineering notes from Andrey Obushev — what we are building, what the work teaches us, and what it takes to make AI trustworthy on real technical work.
Latest post
Why a multi-hundred-page technical document needs more than a good model — and how Corporate AI Portal's Document Translator assembles a glossary while the translation runs.
Read the full entryEarlier entries
