Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableHabr AI→ original

Claude wrote an open-source firmware for an $11 camera in a few hours

The camera, which cost 800 rubles, worked only through a Chinese app and cloud, and no open firmware existed for its chip. The author gave Claude the source code of the OpenBeken project, and the agent dumped the stock firmware over UART, extracted register addresses and 259 sensor settings from it. Within a few hours, the result was a patch with motion detection and photo recording to an SD card. The takeaway: when similar code, a reference sample, and logs are available, porting firmware takes hours.

AI-processed from Habr AI; edited by Hamidun News
Claude wrote an open-source firmware for an $11 camera in a few hours
Source: Habr AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceHabr AI ↗
Editor-in-Chief
BK7252N chip160 MHz ARM, 544 KB of memory, 0.3 MP sensor
Key metric #2An Arduino with a relay power-cycled the camera, and Claude flashed the firmware itself
Everything runs on the local networkno cloud, no registration, and no third-party services

A cheap Wi-Fi camera for 800 rubles typically works only through the manufacturer's app and an account on someone else's server. The author of this story ran into exactly that kind of camera. He wanted to set up video surveillance on his property but did not want to send the footage to a Chinese cloud. No open firmware (the program that lives inside a device and controls its behavior) existed for this model. So he assigned the task to Claude, an AI model from Anthropic that works as an agent with access to the project's files. In a few hours, the agent wrote open firmware that runs without a cloud and can save frames when motion occurs.

What the camera is and why ready-made firmware was not an option

The camera has a magnetic mount, a built-in battery, and USB power. Inside sits a BK7252N chip: a 32-bit ARM microcontroller running at 160 MHz with 544 KB of memory and built-in Wi-Fi. The GalaxyCore GC0329 sensor delivers 0.3 megapixels (648×488), although the store listing advertised HD.

The board had RX and TX test pads. By connecting a USB-UART adapter (a simple converter for a text console) to them, the author saw the RT-Thread operating system. However, no SDK for Beken chips is publicly available. The OpenBeken project, where enthusiasts build open firmware on FreeRTOS, supports similar chips but not this one or this sensor. It is a classic dead end for a hobbyist: no documentation and no ready code.

How the agent got the camera working

Claude got access to the OpenBeken source folder and the task of "getting the camera running." From there, it worked like a careful engineer:

  • it found a UART flashing utility and dumped the original firmware;
  • it built OpenBeken for the closely related BK7231N chip and flashed it to the board;
  • it read the UART logs and saw that the controller starts but the sensor fails to initialize;
  • it determined that the BK7231N has different register addresses controlling the clock for the camera's parallel port, and took the correct addresses from the original firmware;
  • from the same firmware, it extracted values for 259 sensor control registers, after which an image appeared.

The closed loop matters most in practice: build, flash, read the log, fix. To keep the loop from depending on a human, the author found a relay module in an Arduino kit, wrote a simple sketch that accepts commands over USB, and connected the camera's power through the relay. Now Claude could power the board off and on by itself and flash each new version. The human only had to answer in the chat.

What was achieved and the caveats

In a few hours, a patch for OpenBeken with support for this camera appeared. The author also added motion detection: when the frame changes, the camera saves a photo to an SD card. Setup is simple. The camera brings up an open access point with no password, the web interface opens at 192.168.4.1, and under Config → Configure WiFi you enter the home network name and password. After that, the camera can be found on the local network at an address like http://name.local. There is no cloud, no registration, and no third-party services. The code and build instructions are on GitHub, and the author plans Android and iOS apps.

There are limitations too. This is a single hobbyist case on very simple hardware: a 0.3 MP sensor with no sophisticated bootloader protection. The author does not say how many tokens or how much money the experiment consumed, so the economics cannot be assessed. An open access point with no password is appropriate only during initial setup and should not be left that way for long.

Practical takeaway

For embedded developers and IoT product managers, the lesson is this: an agent performs well when three things are present. These are the source code of a similar project, a sample of correct behavior (here, the original firmware), and feedback in the form of logs. If you also automate physical power control, as was done here with an Arduino and a relay costing pocket change, the human drops out of the firmware loop.

Tasks like "port chip support to the neighboring variant" shrink from weeks of work to hours. For hobbyists and small teams, this lowers the barrier to opting out of vendor clouds, and it gives manufacturers of cheap devices reason to rethink tying buyers to their own servers.

Story

Key Questions Answered

What the camera is and why ready-made firmware was not an option?

The camera has a magnetic mount, a built-in battery, and USB power. Inside sits a BK7252N chip: a 32-bit ARM microcontroller running at 160 MHz with 544 KB of memory and built-in Wi-Fi. The GalaxyCore GC0329 sensor delivers 0.3 megapixels (648×488), although the store listing advertised HD. The board had RX and TX test pads. By connecting a USB-UART adapter (a simple converter for a text console) to them, the author saw the RT-Thread operating system. However, no SDK for Beken chips is publicly available.

How the agent got the camera working?

Claude got access to the OpenBeken source folder and the task of "getting the camera running." From there, it worked like a careful engineer: it found a UART flashing utility and dumped the original firmware; it built OpenBeken for the closely related BK7231N chip and flashed it to the board; it read the UART logs and saw that the controller starts but the sensor fails to initialize; it determined that the BK7231N has different register addresses controlling the clock for the camera's parallel…

What was achieved and the caveats?

In a few hours, a patch for OpenBeken with support for this camera appeared. The author also added motion detection: when the frame changes, the camera saves a photo to an SD card. Setup is simple. The camera brings up an open access point with no password, the web interface opens at 192.168.4.1, and under Config → Configure WiFi you enter the home network name and password. After that, the camera can be found on the local network at an address like http://name.local.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…