GhostlyAI is a desktop overlay powered by local AI — it reads your screen in real time, coaches you through interviews, explains code, summarises documents, and answers any question. Zero trace. Zero cloud. Zero compromise.
Free & open source · No sign-up · Your data never leaves your device
WORKS INVISIBLY INSIDE
CORE FEATURES
Renders on a special window layer excluded from all screen capture APIs. OBS, Zoom, Teams, Loom — none of them will see it.
CoreOCR + vision models parse everything on your screen — code, slides, browser, documents — in milliseconds.
Live question detection, STAR-structured answer suggestions, and follow-up prompts — all in real time, invisible to interviewers.
PopularGhostlyAI talks directly to your Ollama instance. Llama 3, Mistral, Gemma, DeepSeek — your choice, your hardware, your rules.
Privacy-firstShow, hide, or switch modes without ever touching your mouse. Works while any window is focused.
Pin a resizable chat window anywhere. It floats above all apps, responds in milliseconds, and disappears from every recording.
Hot-swap between models on the fly. Use a fast model for quick answers, a large model for deep reasoning.
NewGhostlyAI detects what app you're in and adjusts its suggestions automatically — coding help in VS Code, answer coaching in Zoom.
USE CASES
GhostlyAI reads the interviewer's question on screen and suggests structured answers, hints, and code snippets — all without them seeing a thing.
Stuck on a bug mid-call? GhostlyAI reads your editor, identifies the issue, and whispers the fix — instantly, without switching windows.
Paste your slides or notes. GhostlyAI gives you speaker notes, stat checks, and counter-arguments — live while you present.
Open a contract, report, or research paper. Ask GhostlyAI to summarise, find gaps, or draft a response — without leaving the app.
Highlight anything on screen and ask for an explanation. GhostlyAI gives you the depth of a textbook, in seconds.
Ask GhostlyAI to complete repetitive tasks — filling forms, summarising emails, moving files — while you focus on what matters.
HOW IT WORKS
Download Ollama from ollama.com and run it. Pull your preferred model with one command.
Open GhostlyAI — it auto-connects to your Ollama instance. No config, no API keys, no accounts.
Press your hotkey, ask anything about what's on screen. Get answers instantly. Nobody knows it's there.
GhostlyAI runs entirely on your machine using Ollama. No data is sent to any server — ever. Your screen contents, your conversations, your context: all of it stays on your device. No API keys to manage, no subscriptions, no usage logs. Just pure, offline AI.
EARLY FEEDBACK
"I landed my dream job at a FAANG company. GhostlyAI helped me stay calm and structured during the technical rounds. Completely invisible on screen share."
"I use it every day for document reviews. I can ask questions about a contract while on a call with a client and they have no idea. It's wild."
"The local-only approach is what sold me. I'm a security researcher — I can't send screen data to the cloud. This is the only tool that respects that."
FAQ
GhostlyAI uses platform-specific window APIs — WS_EX_TOOLWINDOW and SetWindowDisplayAffinity on Windows, CGWindowListCopyWindowInfo exclusions on macOS — that tell the OS to omit the window from all capture sources. This is the same technique used by DRM-protected video players.
Any model available through Ollama: Llama 3.2, Mistral 7B, Gemma 2, DeepSeek Coder, Phi-3, Qwen, and dozens more. You can also switch models on the fly from within the overlay. No cloud model is ever used.
GhostlyAI itself is lightweight. The main requirement is Ollama and your chosen model. For Llama 3.2 3B you need ~4GB RAM; for larger 7B+ models, 8–16GB is recommended. A dedicated GPU speeds things up but is not required.
That's a fair question. GhostlyAI is a general-purpose desktop productivity tool — no different in principle from having notes open on a second monitor. How you use it is your responsibility. We encourage transparency with interviewers wherever appropriate.
No. Everything runs offline. GhostlyAI communicates only with your local Ollama instance. No telemetry, no analytics, no cloud calls — ever.
Yes. GhostlyAI supports multi-monitor configurations and adapts to any resolution. You can pin the overlay to any monitor independently.
We're in active development on both Windows and macOS builds. Sign up above and we'll send you one email the moment it's ready — no spam, no drip campaigns.
GET NOTIFIED
GhostlyAI is launching soon for Windows and macOS. Drop your email and we'll ping you the moment it ships.