Hold ⌃⌥ to speak, release to capture screen context, and watch answers stream in token-by-token right where you're looking. Zero focus stealing.
maxTokens parameter caps the response generation length at 1024 tokens to keep responses concise and minimize latency.
Hold ⌃⌥ to record audio, release to capture the screen and fire off the request. No clicking or window switching needed.
Speech-to-text uses Apple's native Speech framework locally. Your voice audio never leaves your machine.
Responses stream token-by-token over Server-Sent Events straight into a non-activating floating HUD panel.
Retains the last 12 conversation turns and reuses screen capture context for follow-up questions within 45 seconds.
API tokens are securely stored in the macOS Login Keychain, never written to disk logs or telemetry.
Borderless panel anchored to top-center. Tuck behind the camera notch and expand dynamically on hover.
Triggers global event monitor. Audio engine records level-metered speech locally.
On key release, takes a screen capture under 1568px and runs local transcription.
Model response streams into non-activating overlay anchored at your cursor position.
Take full control of the aesthetic. From the glow and animation of your 3D cursor to the HUD panel's transparency and theme, make InkBuddy match your workspace.
Avoid collisions with your IDE or OS. Fully remap the recording and capture hotkeys to whatever combination fits your muscle memory best.
No locked subscriptions. Plug in your own OpenAI or OpenAI-compatible API keys (via LM Studio, Ollama, etc.) for total transparency and cost control.
Explore the source code on GitHub and build the native macOS companion for your workflow.
Get Source Code