JARVIS

Why it's ranked here

The project connects capture, identification, browser research, synthesis, storage, and live presentation into one coherent pipeline. Its service-aware fallbacks and typed API make the prototype unusually inspectable. However, the repository also describes major pipeline tasks as pending, so the implementation evidence conflicts with its own delivery checklist.

What's good

Each external service can fail or remain unconfigured without stopping the backend. Research can fall back to fast search, storage can fall back to memory, and synthesis can switch providers or return raw findings. Server-Sent Events expose progress as it arrives. Health and service endpoints make missing configuration visible instead of hiding partial operation.

Tradeoffs

Useful results depend heavily on third-party credentials for search, browser automation, identification, synthesis, persistence, and tracing. Running without them preserves the application shell but removes substantial capability. In-memory fallback loses persistence and cross-tab synchronization. The face identification and person-dossier workflow also handles sensitive personal data, while the supplied text describes no consent, access-control, or retention safeguards.

How to use it well

It best suits engineers prototyping a supervised person-research workflow where incremental results and inspectable service status matter. Start with manual image uploads and fast search, then add browser sessions, persistent storage, and synthesis providers deliberately. It does not replace identity verification, privacy governance, source validation, or a general-purpose research system for subjects without faces or names.

Observed

Primary languages

Python backend with a TypeScript Next.js frontend

Packaging and installation

Backend installs as an editable Python package; frontend installs through npm

HTTP interface

FastAPI exposes health, service status, capture, identification, dossier, task, and agent-session endpoints

Streaming interfaces

Research uses Server-Sent Events, and audio transcription uses WebSocket

Input surfaces

Manual uploads, image URLs, video frames, smart-glasses capture, and Telegram photo intake are described

Storage options

Convex provides persistence and real-time synchronization; an in-memory gateway is the fallback

Test surface

Backend pytest commands, frontend Vitest configuration, and Playwright end-to-end configuration are present