Team
Products 2
Bhai
Bhai runs continuously in the cloud with its own computer. It browses, fills forms, reads and sends email, researches the web, edits video and handles reminders, learning your routines and preferences. Passwords stay in a vault it can use but never see. Works on WhatsApp, Telegram, Discord, Slack and desktop.
Bhai is XAGI Labs' always-on personal AI agent. Each user gets a private cloud computer where Bhai can browse the web, fill and submit forms, read and send email, research with live search, work with documents and video, and run long tasks in the background. It remembers people, preferences and past work, and proactively helps only when it has a real reason, such as a deadline or a reply waiting. Safety is built in: logins live in a vault the agent can use but not read, risky actions need the owner's yes, and every task ends with a verified result or an honest stuck report. Bhai talks to you where you already are: WhatsApp, Telegram, Discord, Slack, email and a desktop app, with voice notes and live calls.
https://bhai.xagilab.com/
MELRA
AI agents often report success when nothing changed. MELRA sits between the agent and the world: every action is planned, checked against policy, approved when needed, executed, and confirmed with evidence (file hashes, page contents, HTTP checks). Failures return as partial or unknown, never fake success. Open source.
MELRA is XAGI Labs' open-source runtime that lets AI agents act on real systems without having to be taken on trust. An agent asks what it may do, plans one operation, executes it, and gets a receipt. Policy is enforced at plan and again at execute, with allowed commands and domains and approval for any change. Every change declares evidence up front: file hash, page contains, HTTP resource matches, exit code. Results are verified success, partial, waiting for a human judgement, or recovery required, so a lost reply is never reported as done. Long jobs run as restart-safe workflows. It works with any MCP-compatible agent.