In my endeavour to boil the oceans for accessibility, I've forced Codex to create Private Eye, a continuous (live) description addon for NVDA, using Ollama.
-
In my endeavour to boil the oceans for accessibility, I've forced Codex to create Private Eye, a continuous (live) description addon for NVDA, using Ollama. That means if you have a good enough GPU, you can run Gemma4 or other models on your local machine or AI server. If not, ollama pull gemma4:31b:cloud and you can use Ollama's cloud models. They have a free teer that gives a good hour or so of AI per five hours, and $20 per month which gives 5 times that and I've never hit that limit. Also this is very new. Still, you can edit the general system prompt. It has a general, video, and game description mode. The video and game mode automatically enable when the window title has YouTube, Dizney, Prime Video, ETC. in the tital bar, and game runs when you're using RetroArch, PPSSPP, PCSX2, ETC.
Commands:
* NVDA Alt V: Start and stop. It takes a minute to stop if it's already requesting a description, I'll see if I can get that fixed. Probably see if network requests can use their own thread.
* NVDA Alt Shift V: Clear context.
* NVDA Alt Control V: Open settings.Will I put it on the store? Maybe. I don't feel like it's good enough for that honor yet, but I'll let other people be the judge of this. Note that this is literally like an hour old.
Edit: Github: https://www.github.com/devinprater/private-eye
-
R relay@relay.publicsquare.global shared this topic
-
In my endeavour to boil the oceans for accessibility, I've forced Codex to create Private Eye, a continuous (live) description addon for NVDA, using Ollama. That means if you have a good enough GPU, you can run Gemma4 or other models on your local machine or AI server. If not, ollama pull gemma4:31b:cloud and you can use Ollama's cloud models. They have a free teer that gives a good hour or so of AI per five hours, and $20 per month which gives 5 times that and I've never hit that limit. Also this is very new. Still, you can edit the general system prompt. It has a general, video, and game description mode. The video and game mode automatically enable when the window title has YouTube, Dizney, Prime Video, ETC. in the tital bar, and game runs when you're using RetroArch, PPSSPP, PCSX2, ETC.
Commands:
* NVDA Alt V: Start and stop. It takes a minute to stop if it's already requesting a description, I'll see if I can get that fixed. Probably see if network requests can use their own thread.
* NVDA Alt Shift V: Clear context.
* NVDA Alt Control V: Open settings.Will I put it on the store? Maybe. I don't feel like it's good enough for that honor yet, but I'll let other people be the judge of this. Note that this is literally like an hour old.
Edit: Github: https://www.github.com/devinprater/private-eye
@pixelate I have been waiting for this sort of thing literally for years. This is the real step forward which the models can provide, to begin with.
-
In my endeavour to boil the oceans for accessibility, I've forced Codex to create Private Eye, a continuous (live) description addon for NVDA, using Ollama. That means if you have a good enough GPU, you can run Gemma4 or other models on your local machine or AI server. If not, ollama pull gemma4:31b:cloud and you can use Ollama's cloud models. They have a free teer that gives a good hour or so of AI per five hours, and $20 per month which gives 5 times that and I've never hit that limit. Also this is very new. Still, you can edit the general system prompt. It has a general, video, and game description mode. The video and game mode automatically enable when the window title has YouTube, Dizney, Prime Video, ETC. in the tital bar, and game runs when you're using RetroArch, PPSSPP, PCSX2, ETC.
Commands:
* NVDA Alt V: Start and stop. It takes a minute to stop if it's already requesting a description, I'll see if I can get that fixed. Probably see if network requests can use their own thread.
* NVDA Alt Shift V: Clear context.
* NVDA Alt Control V: Open settings.Will I put it on the store? Maybe. I don't feel like it's good enough for that honor yet, but I'll let other people be the judge of this. Note that this is literally like an hour old.
Edit: Github: https://www.github.com/devinprater/private-eye
@pixelate Brilliant! Do you have any ideas for a local llm that could run on a Hog Alli x alongside the game?
-
@pixelate Brilliant! Do you have any ideas for a local llm that could run on a Hog Alli x alongside the game?
@diegomatos16 I don't know. If nothing else, try gemma4:e2b.
-
@diegomatos16 I don't know. If nothing else, try gemma4:e2b.
@pixelate the pc doesn't have enough hardware
the gemma 4b already uses 14GB. With pcsx2 it reaches almost 16. NVDA was unable to generate any description, unfortunately. -
@pixelate the pc doesn't have enough hardware
the gemma 4b already uses 14GB. With pcsx2 it reaches almost 16. NVDA was unable to generate any description, unfortunately.@diegomatos16 Oh no, try the 2b one. Or Llama3.2.
-
@diegomatos16 Oh no, try the 2b one. Or Llama3.2.
@pixelate That was the one I downloaded, Gemma 4 2B. However, I couldn't really use it because I ran out of RAM. Even with just the addon running, it wasn't able to generate the descriptions properly. With PCSX2 running at the same time, it got even worse.
-
@pixelate That was the one I downloaded, Gemma 4 2B. However, I couldn't really use it because I ran out of RAM. Even with just the addon running, it wasn't able to generate the descriptions properly. With PCSX2 running at the same time, it got even worse.
@diegomatos16 Aww, well, you can always use the Ollama cloud models. gemma4:31b:cloud
