What is avernet?

Avernet is an open source platform built to help many AI agents work together inside an organization, instead of each one operating alone. Think of it as shared infrastructure that lets agents and people connect, coordinate tasks, run their work, and improve over time, all inside one system. It is currently used inside Ant Group, where it reportedly supports over 10,000 agents and bots as of early July 2026. The project starts from a simple observation: as more agents and people join a workflow, four problems tend to show up. Existing capabilities become hard to find, apparent agreement can hide real disagreement, work slows down because a human has to relay information between systems, and knowledge learned along the way does not stick around as something the organization can reuse. Avernet is built to address these four problems directly. To do that, it offers several pieces working together: an identity and permissions layer that handles who can do what and keeps an audit trail, support for running many different kinds of agent engines, containers, and services, a coordination layer where agents can find each other, form teams, and follow shared rules, and a memory and evaluation layer that lets the system keep context and get better over time. On top of these, developers can build agent apps, workflows, and other tools. Getting started involves cloning the repository and choosing one of three setup paths: an automated local install script, a manual setup for people who already have the required tools installed, or a Docker based build that runs everything in containers. None of the paths require a paid AI model key to try the basic features, though connecting a real language model lets the demo bots respond for real. The project is licensed under Apache 2.0 and is still early, described by its authors as community version 0.1, with more of its internal components planned for future release. The full README is longer than what was shown.