Useful work to hand to AI
- Code
- Configuration
- Iteration
- Troubleshooting

A buildable Raspberry Pi camera designed around a Codex-assisted hardware and software workflow.
There is no obvious finished-product equivalent. The point is the combination of custom hardware, image generation and a build process that deliberately gives an AI coding agent meaningful work.
The CAIBI build pack gives an AI assistant the source, the known constraints, the useful AI work and the physical boundary. It is designed to help you move faster without pretending the AI has built the thing for you.
I want to build ImageGenCam. Source project: https://github.com/openai/imagegencam Use the source as the authority. Do not invent parts, dimensions, prices or capabilities that are not published. AI can help with: Code, Configuration, Iteration, Troubleshooting. AI cannot physically do: Print and finish the enclosure, Fit the camera, display, battery and Raspberry Pi into the case, Check buttons, clearances and camera ergonomics in the real device, Verify that the finished camera survives normal handling. Tools and capabilities involved: 3D printer, Raspberry Pi. CAIBI's main concern: The software is only half the product. Enclosure quality, assembly and camera ergonomics still decide whether the result feels finished. Start by checking the source, confirming the current parts and build instructions, then give me a staged plan. Clearly separate source facts from your suggestions. Stop and ask me to verify anything that depends on physical fit, wiring, calibration or safety.
Start with the published Raspberry Pi, display, camera, battery, storage and enclosure parts so the supplied build path matches the hardware in front of you.
Print the supplied camera case. The source says PLA and PETG have been tested and recommends PETG because the enclosure folds.
Open Codex Desktop and give it the ImageGenCam repository. The project is deliberately structured so Codex can guide setup and software work from the source.
Fit the display, camera, Raspberry Pi, battery and storage into the printed shell. This remains physical assembly work.
Bring up the camera, verify capture, prompts, album access, Wi-Fi and the phone companion interface. Codex is useful for configuration and debugging, but you still verify the hardware.
Change prompts, controls, UI, the Magic Button, camera hardware or enclosure. This is the strongest part of the project for AI assisted iteration.
Use the published source, files and instructions as the baseline.
↗AI ASSISTEDUse the CAIBI build packStart from the source with prompts that keep the AI inside the verified build information.
→OUTSOURCE PARTGet fabrication done for youYou do not necessarily need to own every capability. This build uses 3D printer, Raspberry Pi.
→ALTERNATIVE PROJECTClawd MochiA small ESP32-C3 desk companion with animated faces, web controls and a published low-cost parts list.
→When people submit completed builds, this section can link to their build notes, repo, printable files, shop or finished version. CAIBI does not take payment or insert itself into the sale.
Source available and current
There is no obvious finished-product equivalent. The point is the combination of custom hardware, image generation and a build process that deliberately gives an AI coding agent meaningful work.
Actual time, cost, changes and outcomes from reviewed builds will appear here.
Add your build report ↓Your result can improve the assessment and, if you choose, link people to your public build or listing.