Hello.
I'm Witold - a researcher running a number of AI-related experiments and projects. One of them is big, AI-driven, autonomous Haiku port to RPi4. The goal of the experiment was to establish a working and testing environment for an AI agent to create a complete Haiku port to Pi4 autonomously.
The project succeeded and is available at: https://github.com/haiku-pi/haiku-rpi - along with extensive documentation.
The port is fairly complete - with full GPU acceleration (using Mesa 26.2 drivers with OpenGL, OpenGL ES and Vulkan), numerous hardware features, 150+ Haiku packages rebuilt and tested etc. It has some nice and unique features - e.g. fully working Wine 11.20 (current release) supporting running 32- and 64-bit applications and providing x86 / x86-64 compatibility via Box86, Box64 and Fex.
Firefox works with full video playback (H.265 is HW accelerated). Python with numpy work. The system was rebuilt and retested under gcc 16.2. There is even a port of AI-cloned Photoshop called Photcraft (original repo: https://github.com/storytold/photocraft) that was made to work on Pi4 on Haiku with GPU acceleration!
Everything was tested on a real Pi4 device, under an autonomous agentic workflow (agent can hard reboot the device, grab HDMI, control the network environment autonomously thanks to which development and testing was running 24/7).
Along the way, there were also some upstream (non-Pi4 specific) Haiku bugs fixed and a number of performance optimizations delivered. All documented in detail: https://github.com/haiku-pi/haiku-rpi/blob/main/docs/HAIKU-RPI4-CHANGES.md
I know that Renku OS has some Pi4 work already committed, but my analysis shows that "my" port is far more advanced. Full analysis here: https://github.com/haiku-pi/haiku-rpi/blob/main/docs/design/renkuos-analysis.md
When I started my journey, I didn't know: a) about non-AI policy in Haiku; b) existence of Renku OS. Although my goal was purely scientific research (understanding limitations and behavioral patterns of AI agents) and not focused on providing a meaningful open-source contributions to upstream projects - maybe the results of the work can be re-used in purely engineering context.
So if there is an interest within the Renku OS community to re-use any of the products of my work with agents - I'm more than happy to discuss and support this.