For Robotics Engineers

The AI-Native Workbench
Built for Robotics

Let AI do more than write code — let it drive your board directly. Describe a task; the Agent SSHes into your device, runs commands and streams logs back. Chat, terminal, files and flashing — all in one native window. No more tool-switching.

10+ Model providers
7 RDK boards
Native Agent runtime
3 Cross-platform
root@192.168.128.10:22 📶 Dashboard OpenClaw Skills Workshop Terminal 📁 Files 🖥 Remote Desktop </> Code Editor Flasher root@192.168.128… English ▾ ⚙ Settings 💬 Forum Dashboard Workbench No active tasks. Describe a problem or goal to Moss in the box below — progress will show here. ✓ AI Assistant ✓ OpenClaw ✓ Device Online root@192.168.128.10:22 Board · RDK X5 1.1Gi/3.0Gi MEM 54.1°C TEMP 74% CPU 0% BPU 9.6G/29G DISK 3:19 UPTIME ⚡ Quick Dev Terminal OpenClaw Health Check ✓ Dev env ready ✓ IDE ready ✓ Desktop ready Quick ▾ Tell Moss a problem or goal (debug, plan, device action) — or drop a file… About Studio Health YOLO demo
AI Native Blueprint

From cloud brains to the physical edge,
a complete robotics intelligence stack

With Moss as the Agent runtime and RDK Studio as the developer surface, we close the loop “Sense → Reason → Decide → Act” end-to-end.

☁️ CLOUD · LLM PROVIDERS
DeepSeek Qwen GPT Claude Custom API On-device vLLM
Model call · Provider routing
⚡ MOSS · AGENT RUNTIME ENGINE
LLM Provider Router
Auto-failover to backup if primary times out
SessionStore
100k-line terminal scroll, no amnesia
Approval Hook
rm -rf prompts you to confirm first
Knowledge Module
X5 pinout built-in — no forum hunting
Skill Engine · author once, reuse on every board
Device Adapter · zero-cost board migration
CLI / SDK · embed in your own system
3-layer memory · picks up where you left off
UI · LOCAL SSH · ON-DEVICE
💻 RDK STUDIO · DESKTOP
AI Chat Multi-tab terminal File Manager Image Flasher VNC Health Monitor Device Discovery
🔌 RDK HARDWARE · EDGE
RDK X5 · 10T BPU RDK X3 · 5T BPU RDK S100 ROS2 / TROS Cameras / sensors OpenClaw Agent

👨‍💻 Describe the task in natural language → Moss reasons → safety approval → on-device run → logs return → iterate

🔮 ROADMAP: MCP protocol · multi-device cluster orchestration · on-device fine-tuning · voice interaction · ROS topology viz · community skill marketplace · third-party hardware integrations

Workflow Comparison

Same job —
from juggling tools to a single window

Take “get a single camera streaming” as an example, and look at the workflow side-by-side.

Traditional toolchain
Multiple tools
  • Constantly switch between IDE, terminal and flashing utility
  • Copy error logs into a separate AI chat to ask
  • Look up pinouts and camera parameters yourself
  • No unified approval guard for dangerous commands
  • Switching boards means relearning the toolchain
RDK Studio
One workbench
  • All four core capabilities unified in one native workbench
  • AI shares your terminal context — errors are analyzed automatically
  • Pinouts & hardware manuals built-in; ask and get answers
  • Multi-level approval guards; flasher auto-detects USB devices
  • Same workflow across X3 / X5 / S100 — migration is seamless
IDE Experience

AI that edits code — and operates devices

Every capability is deeply integrated with the Agent — it sees your terminal, understands your files, reads your board status, and steps in when you need it.

Chat with Moss

Mid-conversation, the AI quietly SSHes into your board and gets it running

Moss Ready OpenClaw linked Orchestrator · spans Studio, device and OpenClaw + In one line: tell me the IP and board model I'm connected to. Moss · on-device runner (local) · model: qwen3.6-plus | Skills hit: Skill Manager / RDK Board Delegate · 4 total expand bash $ ip addr show | grep 'inet ' | grep -v 127.0.0.1 | awk '{print $2}' ✓ Success 587 ms · device_exec You are connected to a D-Robotics RDK X5 V1.0 developer kit, IP 192.168.128.10, SSH on the default port 22, user root. 11 s · 1 step · 3.1k tokens Tell Moss a problem or goal (debug, plan, device action) — or drop a file… Quick ▾ About Studio Health YOLO demo

Multi-tab remote terminal

Shared context with the chat — the AI can read the exact line that errored

Main + root@192.168.128.10:22 Logs Open Local serial port… 115200 [Connected to Studio backend, spawning PTY…] Welcome to Ubuntu 22.04.5 LTS (GNU/Linux 6.1.83 aarch64) * Documentation: https://help.ubuntu.com * Management: https://landscape.canonical.com * Support: https://ubuntu.com/pro Last login: Sun Apr 19 00:34:07 2026 from 192.168.128.100 root@ubuntu : ~ # ros2 topic list /camera/image_raw /camera/info /cmd_vel /scan /tf root@ubuntu :~# openclaw doctor [doctor] ✓ openclaw v2026.3.28 · gateway listening on 18789 root@ubuntu :~#

Remote file manager

Drag-and-drop transfer, path permissions auto-detected — no more scp

Files root@192.168.128.10:22 ↑ Upload Home / root / .openclaw 🔍 Search this folder… By type ▾ Name Size Modified Actions .cache Apr 18 22:45 Get .config Jan 1 2000 Get .npm-global Apr 18 22:48 Get .openclaw Apr 19 00:33 Get .rdk-studio Apr 18 23:02 Get .bashrc 4053 Dec 8 17:53 Edit Get .bash_logout 220 Jan 7 2022 Edit Get

One-click image flasher

USB devices auto-detected — no separate flashing utility needed

Flash Wizard Pick a device and image, then flash in one click. 1 Pick device Pick image 📥 Flash Done Choose your board Select the RDK developer board you want to flash RDK X3 / X3 Module Entry tier · Sunrise 3 5 TOPS · 4× A53 @1.5GHz 2/4 GB · 200+ open-source algos RDK X5 / X5 Module Mainstream · Sunrise 5 10 TOPS · 8× A55 @1.5GHz 4/8 GB · Type-C flash · WiFi 6 RDK S100(P) XBURN Flagship · Journey 6 128 TOPS · 6× A78AE 12/24 GB · full-stack compute Next → Pick image Learn about boards ▸
CLOUD ↔ EDGE · AGENT PIPELINE LIVE Logs / status / sensor data → AI keeps analyzing AI RDK Studio Moss reasoning STEP 1 Security Hook Intercept · approve STEP 2 X5 OpenClaw · RDK X5 On-device run · return STEP 3 Dispatch task SSH to board
Core Engine

An Agent runtime that truly understands robotics

Generic AI editors focus on “how to write good code.” Moss cares about “how to safely operate a physical device.” From driver crashes and dropped cameras to the last-step confirmation on a rm -rf — every detail is purpose-built for robotics engineering.

Welded into your IDE

A dmesg error in the terminal? The AI hands you a diagnosis on the spot — no copy, no paste, no window switching.

Key differentiator

Cloud-edge dual-agent orchestration

Moss reasons in Studio dispatches commands to OpenClaw on the board logs return AI keeps analyzing. End-to-end loop.

Works offline too

Author a skill once and reuse it across boards. Once dispatched, the board executes on its own — field, weak network or no network, it still works.

Cloud-Edge Architecture

Bridging top LLMs
all the way to physical devices

Moss is the robotics-native Agent runtime built into RDK Studio. It plugs into mainstream LLMs upstream and bridges to the entire RDK lineup downstream via OpenClaw; on the memory side it integrates a 3-layer architecture in collaboration with the Hermes framework, paired with multi-channel orchestration to keep long sessions stable and on-device coordination tight.

MOSS · OPEN ECOSYSTEM v1 Moss · CORE · LLM Large language models OpenClaw On-device Agent Hermes Memory engine · partner ROS2 / TROS Robotics middleware Community Developer ecosystem MCP Protocol bridge

OpenClaw

On-device guardian and gateway

A local-first multi-channel AI gateway. Together with Moss it forms the cloud-edge dual-agent loop.

CORE

Moss

Hardware interception · edge compute adapter · long-log triage

First-party-grade pinout & docs lookup
Multi-level safety approvals
Vendor-neutral hot-swap

Hermes

Long-session memory engine · collaboration

Through our collaboration with the Hermes framework, Moss ships a 3-layer memory architecture (short-term / session / long-term) that survives across sessions and picks up long-running tasks where you left off.

Download today —
and make AI your robotics dev partner

For individual and community developers. Works with the entire RDK lineup — plug-and-play.