gainsec/autoprober

AutoProber is a hardware hacking tool that uses an AI agent to automate the process of probing circuit boards. The problem it solves: when a security researcher wants to analyze the chips and pins on an unknown hardware device, like an IoT gadget, manually locating and physically touching tiny solder pads with a probe is slow and requires constant attention. AutoProber automates this by combining a small CNC machine, a USB microscope, and an oscilloscope into a single agent controlled system. Here is how it works: you place a target device on the CNC platform and tell the agent a new target is ready. The system uses the microscope to scan the board, takes a series of photos, identifies pads, pins, chips, and other features, then stitches those photos into an annotated map. It then presents proposed probe targets on a web dashboard for a human to approve or reject before anything physically touches the board. Once approved, the CNC arm moves the probe to make contact, and the oscilloscope measures the signal. A separate safety monitor checks continuously for unexpected contact, and if anything seems wrong, the machine stops immediately and waits for a human to clear the condition. You would use this if you are doing embedded security research or hardware reverse engineering and want to systematically probe unknown devices without manually guiding the probe by hand. It is written in Python and requires the specific physical hardware described in the project's bill of materials. The safety design is treated as a core part of the project rather than an afterthought, since the system moves physical hardware and could damage a device or itself if something goes wrong. The README explains that the probe pin itself is not trusted as a safety signal, and instead an independent oscilloscope channel is continuously watched during any motion, with any unexpected voltage or alarm forcing an immediate stop and a report to the operator rather than an automatic recovery attempt. All hardware, including the CNC motion, microscope, and oscilloscope, can also be controlled directly through the web dashboard or Python scripts, not only through the agent.