inclusionAI/Avernet

Distributed agent coordination platform where agents live, connect, coordinate, execute, and evolve together — like an organization.

What is Avernet?

Avernet is an open‑source infrastructure layer that lets teams run many AI agents together, keep them coordinated, and let humans interact with them over long periods. It provides the plumbing – identity, authentication, routing, discovery, and governance – so that heterogeneous agents (different runtimes, plugins, or existing bot platforms) can join a single network and collaborate on tasks.

Who might use it?

Enterprises that need dozens or hundreds of bots to work together across business units (the README cites production use at Ant Group).
Developers building multi‑agent applications, such as workflow automation, customer‑service orchestration, or internal knowledge‑base assistants.
Research teams that want a sandbox for experimenting with persistent, coordinated agents without wiring every component from scratch.

Core capabilities (as of the public repo)

Area	What’s available	What’s still coming
Trusted core	Identity, auth (both ✅)	Permissions (partial), security, audit, lifecycle (in progress)
Execution infrastructure	Heterogeneous runtimes, bot‑as‑a‑service, basic containers (✅)	Full container orchestration, clusters (planned)
Agent coordination network	Discovery, relationship building, team formation, routing, collaboration (all ✅)	Governance (planned)
Shared intelligence & evolution	Orchestration, evaluation, evolution tools (AgentEvolve) (✅)	Shared context & memory (planned)
Application building blocks	Canvas UI, workflow engine (TaskGuard) (✅)	Full app marketplace, extensions (planned)

How does it work?

Agents connect via a plugin (e.g., OpenClaw, DeepSeek Harness) or a gateway that forwards tasks to an external bot platform.
The Bot Coordination Service (BCS) – a Rust service – handles registration, routing, and session management.
A frontend workbench (standard or next‑gen) lets users view agents, form teams, and trigger tasks.
AgentEvolve provides diagnostics, benchmarking, and version‑controlled “packs” of bots, enabling continuous improvement.
TaskGuard orchestrates multi‑step workflows across agents.

Getting started quickly

git clone https://github.com/inclusionAI/Avernet.git
cd Avernet
./singlebox/singlebox.sh install-tools   # installs needed binaries
./singlebox/singlebox.sh                # launches a local stack (BCS + UI + 5 test bots)
# Open http://127.0.0.1:8000/ in a browser

For Docker or more complex deployments, see the docs/docker.md guide.

Why does it matter?

Most existing LLM demos focus on a single agent. Real‑world use‑cases—customer‑service suites, internal process automation, or cross‑departmental decision‑making—need many bots that share context, respect permissions, and evolve together. Avernet aims to fill that gap by providing a production‑grade coordination layer that can be extended with any agent runtime.

Current status

The core networking, identity, and coordination features are publicly usable.
Some enterprise‑grade pieces (full permission isolation, audit trails, large‑scale clustering) are still private or in progress.
The project is actively maintained (latest updates listed for Sep 2026) and includes plugins for popular models like DeepSeek.

All information above is taken directly from the repository’s README; no external assumptions have been added.