--- lang: en title: "Nara: no-code physical AI and AI vision platform" url: https://nabrio.com/nara canonical: https://nabrio.com/nara updated: 2026-10-27 publisher: Nabrio Co., Ltd. (nabrio.com) --- # Nara senses the world, decides and acts on it, all on your own hardware. A visual, no-code platform that connects cameras, sensors and machines to AI, and turns what the AI sees into real-world action, running in real time on your own hardware. ## Not just AI models: complete applications Most AI tools stop at a prediction. Nara goes further. In one visual editor you connect your inputs (IP and industrial cameras, sensors, PLCs, web services), add AI models and logic, and drive outputs: reject a product, light a tower lamp, open a gate, send a robot, alert a supervisor, or log the result. You then build a live dashboard for operators on top of it. End users run it as their solution, and integrators and solution builders package it as their own product, extending it with Python, plugins and the HTTP API where they need to. ## What Nara gives each team - **Operations managers**: AI automation that fits their exact process, live dashboards, and data that stays on-site. - **Automation engineers**: Visual flows, direct machine control over Modbus, MQTT, TCP/IP and HTTP, and no code to write or maintain. - **System integrators**: One reusable platform for many customer projects, quick PoCs, and extension through Python and plugins. - **Solution builders and OEMs**: A production-ready foundation for their own physical AI products: flows, dashboards, plugins, an HTTP API and the “Powered by Nabrio” mark. - **IT and security teams**: On-premise deployment, role-based access, scoped API keys, and no required cloud dependency. ## Everything from camera to control signal to dashboard - [Senses, then acts, and connects to almost anything](https://nabrio.com/nara/integrations.md): A flow reads cameras and sensors, interprets the data with AI and rules, and acts in the same system: writes to a PLC, switches a tower light, triggers a robot or AMR, opens a gate, sends a message or stores a record. IP and Basler cameras, Modbus, MQTT, TCP/IP, HTTP, WebSocket, webhooks, Patlite and Primus tower lights, email, Teams and Telegram are built in. - [AI model manager and on-device AI decisions](https://nabrio.com/nara/ai-models.md): More than 60 bundled model families, downloaded, checksum-verified and loaded on demand. Custom models from Reva deploy with no configuration, ONNX models run where compatible nodes exist, and promptable models find objects described in words. The Decision node runs small language and vision-language models locally to answer questions such as “Is the work area tidy?” or “How full is the shelf?” - **Multi-page dashboards**: Build operator interfaces with drag-and-drop widgets: live video with AI overlays, charts, gauges, tables, switches, buttons, inputs, schedule pickers and logs. Widgets can control the flow as well as display it, pages can be split by line, area or role, and an optional public view runs on display screens without login. - **No-code, low-code when you need it**: Drag nodes, connect them and deploy. If/then, switch, loops, counters, delays, variables, arrays, built-in database tables, JSON and XML, maths and text templates, all without code. A Python script node and a plugin system cover the rest, and flows are checked before they deploy so mistakes surface in the editor, not on the line. - **Multi-device sync and role-based access**: Pair several Nara devices into a site and they share person re-identification, known-face galleries and flow variables automatically, in a mesh with no master device and no broker, over encrypted, certificate-based pairing. Five user roles, from Super Admin to view-only Monitor, and scoped API keys with expiry dates. - **High performance, private, and cost-effective**: A native C++ engine with GPU acceleration (NVIDIA CUDA and TensorRT through ONNX Runtime) and multi-threaded flow execution, with trigger queues so bursts of events are never lost. Everything runs on-premise, offline on eligible tiers, with no per-inference cloud fees. ## Easy model deployment from Reva Train a model in Reva, export it as a Nabrio Model (NAM), and drop it into a Nara flow. It needs no configuration. This is the standard path for products, defects and objects that off-the-shelf models don’t recognise. ## Installed and running a first flow in minutes - **Standard installers**: A Windows installer, and packages for Ubuntu, NVIDIA Jetson and Raspberry Pi. - **Starter templates**: Projects that run straight away: webcam detection, a dashboard with triggers, code reading, OCR, pose with zones, face tracking. - **Free courses**: Quick-start guides and free courses (danger-zone monitoring, machine status, conveyor parcel processing) on help.nabrio.com, in English and Thai. - **A Free tier**: Evaluate with up to 4 hours of deployment and unlimited development time per day. ## Scales from an edge device to a GPU server Minimum: 1 GHz quad-core CPU, 2 GB RAM, 8 GB storage. Eight CPU cores are recommended for busy flows. A GPU is recommended for multiple camera streams and heavier models. | OS | Version | Architecture | GPU (CUDA) | | --- | --- | --- | --- | | Windows | Windows 11 | x86_64 | Supported | | Linux | Ubuntu 22.04 / 24.04 | x86_64 | Supported | | Linux | NVIDIA JetPack 6.2 (Jetson) | aarch64 | Supported | | Linux | NVIDIA JetPack 7.2 (Jetson Thor) | aarch64 | Supported | | Linux | Raspberry Pi OS (64-bit) | aarch64 | CPU only | ## Ready for AI agents - **Agent interface (CLI and MCP)** (Upcoming) Today: on-device AI decisions, promptable detection and segmentation, Python scripting, and a full HTTP API with scoped keys that scripts and agents can use. Coming next, the Nara agent interface (CLI and MCP server) will let AI agents build a flow and its dashboard from a plain-language request, validate it, deploy it, read its live results and improve it, with permission presets, per-device policies, confirmation before risky actions, simulated I/O, a sandbox, rollback and an audit trail. ## Frequently asked questions ### Is Nara production-ready? Nara 2.0 is in public beta, with features and UI largely stable. Nara has run in production since March 2021, starting with real-time contamination QC on 455 tonnes of fertiliser pellets a day. ### How is Nara different from n8n or Node-RED? n8n automates business software. Nara automates the physical world: cameras, sensors, machines and robots, in real time, with 60+ AI model families running locally. They work together: Nara can call n8n over HTTP or webhooks and receive triggers from it. ### Can we extend Nara ourselves? Yes. Add custom logic with the Python script node, new nodes with the plugin system, and control Nara from your own software through the HTTP API.