Pluely – a privacy first AI assistant (github.com)

Pluely is an open-source, privacy-first desktop AI assistant that replicates Cluely’s real-time meeting/interview features but runs locally and is dramatically lighter — ~10MB vs ~270MB, with <100ms startup and ~50% less CPU/RAM usage. It offers a translucent always-on overlay, global keyboard shortcuts (voice, screenshot, toggle), eight specialized modes (meetings, interviews, presentations, coaching, translation, etc.), continuous system-audio capture with smart context detection, and instant speech-to-text powered by advanced VAD. Conversations, screenshots, and histories are stored locally (exportable to Markdown), screenshots can be auto- or manually submitted for AI analysis, and users can hide the app icon for stealth operation. For the AI/ML community and privacy-conscious teams this matters because Pluely unbundles real-time assistant UX from centralized data collection and vendor lock-in: it supports any LLM or STT provider (OpenAI, Gemini, Claude, Groq, Mistral, Cohere, Ollama, Whisper, ElevenLabs, Google, Deepgram, etc.) and adds curl-based custom provider integration with streaming/non-streaming support, configurable response paths, auth headers, and templated variables ({{TEXT}}, {{IMAGE}}, {{SYSTEM_PROMPT}}, {{MODEL}}, {{API_KEY}}). This makes low-latency, local or private-cloud deployments feasible on commodity machines and enables reproducible, auditable assistant pipelines — useful for onsite deployments, research, and teams that need full control over models, data flow, and compliance.