AutoProber: AI-driven flying probe automation for hardware hacking
What AutoProber does
AutoProber is a hardware hacker's flying probe automation stack designed to give AI agents everything needed to go from detecting a new target on a plate to safely probing individual pins. The system handles target discovery, microscope mapping, safety-monitored CNC motion, probe review, and controlled pin probing.
Workflow and operation
The typical workflow involves telling the agent to ingest the project, connecting all hardware, having the agent confirm all parts are functioning, running homing and calibration, attaching the custom probe and microscope header, and then notifying the agent of a new target on the plate. The agent will find the target location, take individual frames while recording XYZ coordinates, note pads, pins, chips, and other features, stitch frames together, annotate the map, add probe targets to the web dashboard for approval or denial, probe approved targets, and report back.
Hardware control and safety
All hardware can be controlled through the web dashboard, Python scripts, or by the agent itself. The project treats hardware movement as a machine-control system rather than a normal web app. The safety model requires continuous monitoring of oscilloscope Channel 4 during any motion, with any Channel 4 trigger, ambiguous voltage, CNC alarm, or real X/Y/Z limit pin serving as a stop condition. Recovery motion is not automatic.
Repository structure
apps/- Operator-facing scripts and Flask dashboard entrypointautoprober/- Reusable Python package for CNC, scope, microscope, logging, safetydashboard/- Single-page web dashboarddocs/- Architecture, device references, operations, and safety guidancecad/- Printable STL files for the current custom toolheadconfig/- Example environment/configuration files
Hardware stack
The tested architecture uses:
- GRBL-compatible 3018-style CNC controller over USB serial
- USB microscope served by mjpg_streamer
- Siglent oscilloscope over LAN/SCPI for Channel 4 safety monitoring and Channel 1 measurement
- Optical endstop wired to an external 5V supply and oscilloscope Channel 4
- Optional network-controlled outlet for lab power control
Reference parts
The prototype uses specific parts including optical end stop, USB microscope, SainSmart Genmitsu 3018-PROVer V2, Matter Smart Power Strip, Siglent SDS1104X-E Oscilloscope, Dupont wires, pen spring or similar light compression spring, and 3D printer for printable toolhead parts.
License and status
The repository is a self-contained source-available release candidate under PolyForm Noncommercial 1.0.0 license with commercial contact available. It uses Python with dependency resolution via uv.lock.
📖 Read the full source: HN AI Agents
👀 See Also
Spec-Driven Development Workflow for Claude Code: Decomposition, Context Clearing, and Cost Control
A spec-driven development approach for Claude Code that uses two-dimensional decomposition, context clearing between steps, and specs written to disk to improve agent performance and reduce costs.
Fable 5 in Claude Code: Day One Cost Analysis — $210 API-equivalent, $0 Paid
A developer switched to claude-fable-5 in Claude Code and measured token usage across 742 replies. API-equivalent cost: $210.15. Actual paid: $0 during the plan window until June 22.
Users Report Mixed Value from OpenClaw and ClawDBot: What You Need to Know
OpenClaw and ClawDBot, while promising AI tools for code automation, have left some users underwhelmed. This article explores key insights from a Reddit discussion on user experiences and value derived from these platforms.
Local LLM Performance Benchmarks on Mac Mini with OpenClaw and LM Studio
A Reddit user posted performance figures for running the Unsloth gpt-oss-20b-Q4_K_S.gguf model locally on a Mac Mini with 32GB RAM, achieving 34 tokens/second with a 0.7 second time to first token using OpenClaw 2026.3.8 and LM Studio 0.4.6+1.