GhostDesk
Real-time AI overlay for meetings & invisible to screenshare
106 followers
Real-time AI overlay for meetings & invisible to screenshare
106 followers
GhostDesk is a free AI overlay assistant for Windows that floats above any app and stays invisible during screen shares, Zoom calls, and recordings. No tab switching, no distraction — powered by Llama and GPT-4.1.
@mit_parikh Congratulations on this! I really resonate with the no tab switching. Curious how GhostDesk handles heavy accents or fast speakers? Would love to know what your real-world WER looks like in noisy call environments.
The Deepgram Nova-3 + GPT-4.1 + Llama 4 Scout pipeline is a clever architecture choice - real-time transcription needs that low-latency local model, while the reasoning layer benefits from the larger context window.
What really stands out is solving the alt-tab anxiety problem. I've been in interviews where I desperately wanted to reference notes but didn't want to look distracted - this solves that elegantly.
Question: How does it handle accent diversity and technical jargon in transcription? And is there any plan for a Mac version?
The 'invisible to screenshare' feature isn't just a nice-to-have in certain markets; it's a cultural necessity. In MENA enterprise environments, particularly GCC countries, there's an unspoken professional standard where displaying AI assistance tools during client meetings can signal inexperience or lack of mastery. I've seen sales teams and consultants manually turn off Otter/Fireflies before screensharing despite needing the support. GhostDesk's architecture solves a legitimately different problem than standard meeting AI tools.
Congrats @mit_parikh I appreciate the ease and accessibility of this! Does GhostDesk support custom knowledge bases? For example, could I feed it my product docs, or research notes so it surfaces my specific answers rather than generic GPT responses?
The 'invisible to screenshare' feature is genius - solves the awkward alt-tab moment everyone has experienced in interviews. Smart choice pairing Deepgram Nova-3 for transcription with GPT-4.1 for context. Two questions: How do you handle the latency between transcription and AI response for real-time use cases? And does the overlay work with virtual camera setups like OBS virtual cam for streaming? This could be huge for content creators doing live interviews where they need discreet prompts.
As someone who's had to awkwardly pause screenshares to search docs mid-call, the invisible overlay concept is brilliant. Being able to silently pull up relevant client data or talking points without breaking flow would save me from those "let me just..." moments that kill momentum.