--- lang: en title: "The physical AI platform: Nara and Reva" url: https://nabrio.com/platform canonical: https://nabrio.com/platform updated: 2026-10-27 publisher: Nabrio Co., Ltd. (nabrio.com) --- # The platform for physical AI. Use it as is, or build on it. Nabrio connects cameras, sensors and machines to AI models and turns what those models see into actions: rejecting a defective product, opening a gate, alerting a supervisor or signalling a robot. With Nara, teams build complete applications visually and run them in real time. With Reva, they train their own models without code. Everything runs on the customer’s own infrastructure. ## What we provide, and what stays yours - **Applications** — What people use: Finished solutions run by end users, and products built and resold by integrators and OEMs, co-branded “Powered by Nabrio”. - **Nabrio platform** — What we provide: Nara: cameras and sensors, 60+ AI model families, logic and rules, PLCs, robots, MQTT and APIs, dashboards. Reva: custom models. - **Your infrastructure** — Where it runs: IP cameras, sensors, PLCs and machines, edge devices and PCs, GPU servers, and your own cloud servers. - **Your systems and third-party apps** — Alongside: MES, SCADA and ERP; Teams, Telegram and email; n8n, Power Automate, Home Assistant, Grafana; cloud services, over MQTT, HTTP, WebSocket and APIs. ## From data to model to action Reva and Nara are designed together. A model trained in Reva deploys to Nara as a Nabrio Model (NAM) file with no extra configuration, and new images from production go back to Reva to retrain it. - [Reva: train](https://nabrio.com/reva.md): Manage datasets, annotate and auto-annotate, train and evaluate. No code, and data stays on-premise. - [Nara: run](https://nabrio.com/nara.md): Inputs from cameras, sensors and devices; AI and logic in a flow of nodes; actions on machines, alerts and data; live dashboards. ## Six steps from camera to control signal - **Connect**: Add the cameras, sensors and machines to a Nara flow. - **Detect**: Use a bundled model (people, faces, PPE, weapons, vehicles, fire, licence plates, text and more), or train a custom one in Reva. - **Decide**: Add rules, zones, counters, timers and AI decision nodes. - **Act**: Drive PLC registers, tower lights, gates, robots, MQTT topics, emails, Teams and Telegram, and log data. - **Show**: Build a live, multi-page dashboard for operators and managers. - **Improve**: Collect new images from production, retrain in Reva and redeploy. ## A solution for end users, a foundation for builders | | Use it as a solution | Build on it | | --- | --- | --- | | Who | End users: manufacturers, retailers, operators of buildings and cities | System integrators, solution builders, hardware vendors and OEMs, in-house engineering teams | | What they do | Start from built-in nodes, bundled models and project templates, set up zones, rules and I/O, and run it | Build their own applications and products on Nara and Reva, then package, resell and support them, co-branded “Powered by Nabrio” | | What they get | A working AI system in days, without code | A production-ready foundation, so their engineering goes into their product, not the infrastructure under it | Both run on the same platform, so a customer can start with a ready solution and extend it later, and a partner’s product gets every platform improvement. ## Everything a physical AI application needs - **Flows of nodes**: Application logic built visually from tested nodes: inputs, AI models, logic and outputs. - **Multi-page dashboards**: The application’s user interface: live video, charts, controls and set-points, with an optional public view for display screens. - **Python script node**: Custom logic inside a flow, where no node exists. - **Plugin system**: New nodes for your own devices, algorithms or integrations. - **HTTP API with scoped API keys**: Control and data access from your own software, portals and backends. - **HTTP, webhooks, MQTT, WebSocket**: Connection to any outside system: MES/SCADA, ERP, n8n, cloud services. - **Reva custom models and training plugins**: Models for your own objects and defects, deployed to Nara as NAM files, and new training engines through MCP-based plugins. - **Role-based access and multi-device sync**: Applications that serve many users and cover a whole site. - **“Powered by Nabrio” mark**: Co-branding for partner and OEM products. - **Agent interface (CLI and MCP)** (Upcoming): AI agents that build, deploy and improve flows with the same validated nodes. ## Starting with vision, expanding to every kind of data Nabrio started with AI vision and is expanding to every kind of data a physical system produces: sound, text and sensor signals, so machines can understand their environment as a whole and act on it. | Domain | Today | Next | | --- | --- | --- | | Vision | Detection, classification, tracking, faces, pose, actions, OCR, codes, segmentation; custom training in Reva | More model types and training plugins | | Text and language | Small on-device language and vision-language models (Decider, OmniJev) answer questions about text, images and video in a flow | Language and LLM fine-tuning in Reva | | Time series and sensors | Read sensors and machines over Modbus, MQTT, HTTP and TCP/IP, with logic, maths and moving-window statistics | Anomaly detection, forecasting and predictive maintenance models trained in Reva | | Speech and audio | Not yet | Audio classification and speech recognition | ## Your hardware, your data Nara and Reva run on customer-owned infrastructure: on-premise, at the edge, or on your own cloud servers, and offline where needed. The Nabrio Cloud Portal, coming soon, will handle accounts, organisations, subscriptions, slots, offline licences and downloads. It manages entitlements only and does not process customer data or run the products.