PVYai
Screenshots
Overview
PVYai is a comprehensive, PriVacY-protecting AI platform with 25+ specialized model presets, cross-session memory, predictive action validation (PVYai Sentinel), RAG decision management, and 248+ native Function Calling MCP tools — all hosted on Swiss sovereign cloud infrastructure with zero data leakage.
PVYAI enhances PVYoffice, PVYbusiness OS, PVYsearch, PVYservice Desk, PVYvault (with AI-specific Vaults only), and PVYautomat, with more integrations planned. Users receive automatically provisioned OpenCode Desktop and VScodium/VSCode plugins tailored to their preferences.
Hosted on your own sealed PVY@Cloud environment or PVY-Appliance with PVY-ID single sign-on or as Hybrid Variant.
PVYai provides a personalized, enterprise-grade AI platform of the highest caliber.
Equipped with large context models (up to 1 Million tokens), fast inference (up to 370 tok/s), and 248+ MCP tools exposed as native function calls — no special MCP client or protocol required by the LLM. It features PVYai Memory for institutional knowledge that survives team changes, PVYai Sentinel for predictive action validation across 7 domains, RAG decision management with BGE-M3 multilingual embeddings, and native tool calling integrated into the entire PVYapps ecosystem. From coding and DevOps to research, translation, image generation, and business automation — PVYai delivers the right AI for the right task, automatically, with enterprise governance and Swiss PriVacY by design.
Features
- 25+ Specialized AI Model Presets — Coding, DevOps, Research, Translation (55 languages), Image Generation (Photo-realistic, Product Mockup, Illustrations, Diagramms, Aquarels etc) Document Converting (PDF, DOCX, PPTX, XLSX to Markdown), UX Design, Architecture, Analytics, and more — each with tailored system prompts, context windows, and tool access
- MCP Tools as Native Function Calling — All 248+ MCP tools (GitHub, GitLab, Web Search, Playwright, Draw.io, PDF2MD, PostgreSQL, Docker, Memory, Filesystem, Time, SQLite, and more) are exposed to every PVYai LLM and Agent as standard OpenAI-compatible function calls. No MCP client, no special protocol, no extra hop — the middleware transparently resolves and executes them server-side. Any FC-capable model gets the full tool arsenal natively
- PVYai Memory — Cross-session knowledge preservation with semantic search, knowledge graph with temporal validity, wing-based access control (primary/secondary/tertiary groups + private wings), drawers mined from infrastructure configs, and interactive settings panels for admins and users
- PVYai Sentinel — Predictive action validation simulating outcomes before execution across 7 domains (Terminal, SWE, MCP, OS, Search, Web, Android). Detects destructive commands, predicts code change side effects, and triggers session handovers at 70% context threshold with adaptive summaries (250-400 tokens).
- PVYai Agent — All PVYai Features and more in your favorite Shell on macOS, Linux or Windows PowerShell (also on Embedded Systems) Only 33 MB in size, you can define sandboxed or open, interactive or fully autonomous Operations, supports Loop and Swarm Design.
- Context up to 1 Million Tokens — Handles long-form research, repo-scale refactoring, full document analysis, and multi-day sessions with models like DeepSeek V4 Pro (1M ctx) and GLM-5.2 (976K ctx)
- Token Caching — 90-95% savings on prompt tokens via Anthropic native prompt caching with ephemeral cache_control markers. AGENTS.md, skills, and system prompts cached after first request
- PVYtranslate — Word document translation in 55 languages (WMT25 MQM 5.86+), preserving all formatting, fonts, styles, tables, headers, and footers. EN↔DE primary with optional post-editing for grammar and style
- PVYai Art Director — Photorealistic image generation with physics-aware prompting (human anatomy, quadruped anatomy, fire physics, scene detection), built-in inpaint editor (Flux.1 Fill), 4 LoRAs, and automatic scene-specific guidance. Powered by Flux.1 Dev on dedicated GPU
- PVYoffice Integration — Contextual AI in documents: translate, rewrite, summarize, shorten, image generation, spreadsheet analysis. Available via OnlyOffice plugin with b64_json inline image support
- PVYbusiness OS Integration — AI-driven CRM/ERP: lead enrichment, project proposals, dashboard insights, BPM automation. 4 Directus AI models for research, architecture, translation, and chat
- OpenCode Desktop & VSCode Integration — 13 coding models with native tooling, sub-agent orchestration, up to 370 tok/s inference, vision support for screenshot analysis, and thinking preservation for reasoning models
- PVYai Teams & Sandboxes — Autonomous 24/7 AI teams via PVYlima Desktop: Dev Team, Red Team, Sales Team, Support Team — each with dedicated model presets, tool access, and memory wings
- RAG Decision Management — Retrieval-augmented generation with multilingual embeddings (100+ languages, 8K context), ChromaDB vector database, PVYsearch meta-search (10 engines), and citation-formatted responses
- Corporate Governance & Compliance — Granular user/group permissions, AI declaration policies, customizable PDF reports with branding, PII stripping filter (Presidio) for cloud models, and PVYai PriVacY Filter for enterprise data protection
- PriVacY by Design — All data stays in Swiss datacenter. Hermetically sealed per customer. PII stripping for cloud API models. No data leakage to third parties. Firewall-restricted GPU infrastructure with IP whitelisting
- OpenAI-Compatible API — Connect any app, service, or custom tool via standard
/v1/chat/completionsendpoint. Full SSE streaming, function calling, vision, and reasoning_content support
Highlights
- Swiss Sovereign AI — PriVacY-first, hermetically sealed per customer, all data stays in the Swiss datacenter. No third-party data sharing
- MCP as Native FC — Unlike traditional MCP setups requiring special clients and protocol awareness, PVYai exposes all 248+ MCP tools as standard OpenAI function calls. Every model gets the full tool arsenal natively — no extra infrastructure, no protocol translation, no model retraining
- PVYai Sentinel — Predictive, not reactive: simulates command and code change outcomes before execution, prevents breaking changes, and triggers intelligent session handovers at context thresholds
- Cross-Session Memory — Institutional knowledge that survives team changes, with wing-based access control, knowledge graph with temporal validity, and interactive admin/user settings panels
- Model Orchestration — The right AI for the right task, automatically. 25+ specialized presets with tailored system prompts, context windows, tool access, and thinking modes
- 90-95% Token Savings — Anthropic native prompt caching for AGENTS.md, skills, and system prompts. Cache hit on every subsequent request
- Enterprise Governance — Granular permissions, compliance declarations, customizable branding, PII filtering, and AI declaration policies
BTW; Its pretty cool if your PVYai files a PVYai Shared Memory into Drawer from within PVYbusiness OS about new Data Fields where being created, a second later you recall this Note from OpenCode Desktop and let the same or a different PVYai Model implement the new Data Fields in your frontend automatic, respecting all SSR Rules, re-deployes you a new container to test, and before you have time to open your Web Browser you have the Feature and Security Test Report already on your screen.
Privacy Guaranteed
Like all PVYapps, PVYai sends zero telemetry data. Your usage patterns, content, and metadata stay completely private within your sovereign cloud.
Get Started
PVYai is included with every PVY.swiss cloud deployment. Start using it today.
Deployment Options
- SaaS on your domain
- On-Premise available
- Swiss data center
App Information
- Current release
- v0.9.9Stable
- Platform availability
- WebMacOSLinuxWindows
- Planned availability
- IOSAndroid
- Available on
- PVY@Cloud SaaS @businessPVY@Cloud SaaS @enterprisePVY@Office Appliance
- Capabilities
- Third Party Extensions