(formerly nRF AI Debugger)
AI debugging agent for IoT SoCs, captures live logs from your connected nRF devices, analyzes application behavior, and generates expert insights — right from VS Code.
Debugging firmware on nRF devices is notoriously tedious. You flash your firmware, open a terminal, and watch raw logs (RTT/UART) scroll past. Trying to correlate timestamps between two boards, decipher hex codes, and manually search your source code to find where an error originated is a major productivity bottleneck.
SoC AI Debugger changes that. It's an AI agent built specifically for the nRF Connect SDK ecosystem. It captures live logs directly from your boards and analyzes them in real-time, correlating firmware output with your source code to pinpoint the root cause of failures.
The SoC AI Debugger captures live RTT or UART logs, identifies patterns in your application's behavior, and produces structured analysis reports—covering everything from boot sequences to protocol-specific events.
Key Capabilities:
- 🔍 Auto-detects connected boards via J-Link serial numbers.
- 📡 Multi-device capture — two devices (or more) simultaneously (e.g. Central + Peripheral).
- 🧠 Context-aware analysis — correlates logs with your actual
main.cand project files. - 💡 Proactive Debugging — catches Hard Faults or stack overflows and points to the offending line of code.
Before you can analyze, you need good logs. The agent reads your nRF Connect SDK project, understands the BLE stack, and injects the right log statements — so when it analyzes later, it knows exactly what each line means.
Key Capabilities:
- 📁 Multi-project awareness — handles Central + Peripheral workspaces simultaneously
- ⚙️ Auto-configures logging backend (RTT vs UART) in
prj.conf - 🎯 NCS-compliant — follows Zephyr RTOS logging best practices
- 🔘 Interactive — asks before modifying, suggests RTT over UART for BLE projects
Why does the agent generate the logging code? Because an agent that wrote the log statements can analyze the output far more intelligently — it understands the context because it created it.
- Install SoC AI Debugger from the VS Code Marketplace.
- Configure your AI provider (We recommend GLM-4.7 for cost-effective, high-performance analysis).
- Choose a mode: "Analyze nRF Device Logs" or "Generate Logging Code".
| Requirement | Details |
|---|---|
| nRF Connect SDK | Tested with v3.2.1 |
| Extension Pack | Requires nRF Connect Extension Pack |
| Python | 3.8+ (Uses the Python environment bundled with your nRF Connect Extension) |
| AI Provider | Supports OpenRouter or any OpenAI-compatible endpoint. |
We are expanding based on community needs. If you need support for a specific protocol or board, join our discussions!
| Category | Supported / Tested | Future Exploration (User Driven) |
|---|---|---|
| Boards | nRF52840 DK, nRF52832 DK | nRF53, nRF91, nRF70, nRF54 |
| Protocols | BLE (Bluetooth Low Energy) | WIFI, Thread, Matter, LTE-M / NB-IoT, DECT NR+ |
| NCS Version | v3.2.x | v2.9.x LTS, v3.3+ |
| LLMs | GLM-4.7, Claude Haiku 4.5 | DeepSeek-V3, Local LLMs (Ollama) |
| Model | Provider / Endpoint | Status | Notes |
|---|---|---|---|
| GLM-4.7 | OpenAI-Compatible (Coding Plan) | ✅ Recommended | Best balance of high coding benchmarks and extreme cost-effectiveness. |
| Claude Haiku 4.5 | OpenRouter | ✅ Tested | The fastest and most affordable entry-point for professional-grade coding models. |
"We optimized for these models so you can debug for hours for the price of a cup of coffee."
New to GLM? Follow the Step-by-Step Configuration Guide to get your API key and set up the OpenAI-compatible endpoint in VS Code.
Your firmware stays yours.
- Local Control: The agent runs entirely on your machine. It only sends specific log snippets and code context to your chosen AI provider.
- BYOK (Bring Your Own Key): You have full control over which model you use and which API endpoint you trust.
- Open Source: Our capture scripts and agent logic are fully transparent and auditable by the community.
nRF AI Debugger collects basic, anonymous usage data to help us understand which features are most valuable and catch silent errors (like missing python packages or unsupported OS commands) before you even have to report them on GitHub.
What we track:
- Extension activations (to understand daily usage).
- Which logger tools you trigger (e.g.
uart-loggervsrtt-logger). - Tool execution crashes or errors.
What we DO NOT track:
- We NEVER collect your source code, your absolute file paths, or anything typed into the interactive AI chat.
- We NEVER collect the raw logs coming off your firmware devices. This data is strictly private and sent only to your configured AI provider.
Opt-Out: We respect VS Code's global telemetry settings. If you wish to disable this tracking, simply set
telemetry.telemetryLeveltooffin your VS Code settings.
graph LR
A[Your nRF Project] -->|Agent reads code| B[Log Generator]
B -->|Injects log statements| A
A -->|Build & Flash| C[nRF Device]
C -->|Live RTT/UART capture| D[Log Analyzer]
D -->|Code-aware analysis| E[Expert Report]
Adsum Networks — We've been developing IoT solutions on nRF and other embedded platforms for over 7 years. We built nRF AI Debugger because we needed it ourselves to handle complex BLE debugging, and now we're sharing it to help the community build better firmware, faster.
This project is an independent, community-developed tool and is not affiliated with, endorsed by, or sponsored by Nordic Semiconductor ASA.
"nRF" is a registered trademark of Nordic Semiconductor ASA. All other trademarks are the property of their respective owners.
- Cline — The open-source AI assistant this project builds upon.
- Nordic Semiconductor — For the exceptional nRF Connect SDK and developer tools.