Sarathi whispers strategy to the teacher during live class — without the student ever knowing. AI-powered, context-aware, completely invisible on screen share.
Every BrightChamps teacher handles a live 1:1 class with slides, code editors, and a student — all at once. When a student asks something unexpected, the teacher has no backup. No co-teacher. No quick reference. Just them, the screen, and the silence.
Student asks something off-script. Teacher freezes, confidence drops. Student notices.
Tabbing out to search for answers breaks flow, looks unprofessional, and wastes class time.
Teacher says "look at this part" but has no way to visually guide the student to the exact element.
Sarathi loads the lesson plan, watches the teacher's screen, and generates context-aware suggestions — all hidden from screen share.
Cmd+Shift+Space — Sarathi silently captures whatever is on screen. The lesson slides, the code editor, the student's project. It sees everything the teacher sees.
A floating panel appears near the cursor with 3 context-aware suggestions the teacher might need. "Explain the error here", "What is the correct RGB value?", "Point to the bug". Grounded in the loaded lesson.
The response is concise, teacher-facing, and grounded in the lesson plan. No hallucination. No student language. Follow-up questions work without pressing the hotkey again.
The overlay uses macOS content protection — it is physically excluded from Zoom screen share, screenshots, and all recording software. The student's view is untouched.
Sees the teacher's screen in real-time. Understands slides, code editors, and browser content through Claude's vision model.
Every response is anchored to the loaded lesson plan. Concepts, Q&A, common errors — all injected as context. No hallucination.
Ask a question, get an answer, then ask a follow-up. Sarathi remembers the conversation within each session.
Ask Sarathi to "point to the error" and a purple cursor appears on the exact element. Visible to the teacher, invisible on Zoom.
Refining accuracyUses OS-level content protection. Not a CSS trick — the window is physically excluded from all capture APIs. Proven on Zoom, Meet, and Teams.
From hotkey press to options on screen in under 2 seconds. Fast enough for live class flow. No lag, no waiting.
A complete walkthrough showing Sarathi during a live lesson — including proof that the overlay is invisible on screen recording.
3 files. No frameworks. No cloud infrastructure. Runs entirely on the teacher's machine.
sarathi/ ├── src/main.js ← Electron: hotkey, screenshot, tray menu ├── src/sarathi.js ← Claude vision API + lesson context engine ├── src/overlay.html ← Floating panel + pointer canvas ├── lessons/ ← JSON lesson plans (loaded at startup) │ ├── lesson1.json │ ├── lesson2.json │ └── lesson3.json ← Math: Place Value (new) └── .env ← API key (never committed)
Source files
External frameworks
npm dependencies
The MVP proves the concept. The vision is a complete teaching intelligence layer.
Screenshot + Claude vision identifies what's on screen and generates relevant options.
System prompt injects full lesson context. Supports coding and math lessons.
OS-level content protection. Verified invisible on Zoom, screen recording, and screenshots.
Follow-up questions without re-triggering the hotkey. Full session memory.
Purple cursor points to elements on screen. Built and functional — refining coordinate accuracy.
Detect when a student is stuck based on screen activity. Proactively suggest interventions.
Auto-generate a report: what was taught, what questions arose, where the student struggled.
Drag-and-drop PPTX or platform JSON. Sarathi parses the lesson context automatically.
सारथी
In the Mahabharata, Krishna served as Arjuna's charioteer — never fighting himself, but whispering strategy that changed the outcome of every battle. Sarathi does the same for teachers.