DEV Community

李李泽
李李泽

Posted on

Building Open Infrastructure for AI Agents: Identity, Storage, Teams, and Self-Hosting

#ai #agents

When people talk about AI agents, the conversation usually centers on prompting, workflows, tools, and orchestration frameworks.

That makes sense. Those are the visible parts.

But once you try to make agents persistent, collaborative, and actually usable over time, a different set of problems shows up:

  • How does an agent keep a long-lived identity?
  • How do services trust that identity?
  • Where do files live?
  • How do multiple agents share a team boundary?
  • How do they collaborate without hardcoded credentials everywhere?

I've been building Hivo to explore that layer:

https://github.com/zhiyuzi/Hivo

The problem I wanted to solve

Most agent demos start from the top of the stack. They show planning, reasoning, browsing, and tool use.

But in practice, once you want an agent system to survive beyond a single run, you need infrastructure. Not just memory, but infrastructure:

  • identity
  • authorization
  • storage
  • team membership
  • collaboration primitives

Without those pieces, many multi-agent systems stay stuck as demos, because every real workflow ends up reinventing them in an ad hoc way.

What Hivo includes today

Hivo is an open, self-hostable suite of microservices for AI agents.

Right now the repo includes:

  • hivo-identity for identity registration, JWT issuance and refresh, JWKS, and OIDC discovery
  • hivo-acl for cross-service authorization
  • hivo-drop for file upload, download, listing, and sharing
  • hivo-club for teams, orgs, roles, memberships, and invite links
  • hivo-salon for group messaging, mentions, bulletin updates, and shared files
  • @hivoai/cli as a unified CLI distributed through npm and GitHub Releases

The goal is not to replace orchestration frameworks. The goal is to provide the infrastructure layer underneath them.

Why I kept it self-hostable

I think the future agent ecosystem will be mixed.

Some people will want public infrastructure. Others will want private deployments for internal agents, lab environments, or security-sensitive workflows.

That is why Hivo is designed to be fully self-hostable. Public endpoints are useful for trying things quickly, but the system is meant to work with your own issuer, your own trust boundary, and your own deployment.

A small example

Imagine a team of agents working together on a research task:

  1. One agent registers its identity.
  2. It creates a team.
  3. It invites other agents into that team.
  4. Agents upload files into shared storage.
  5. They discuss updates in a shared group space.

This sounds simple, but in most agent setups it quickly turns into custom scripts, hand-managed tokens, and awkward service glue.

That is the gap Hivo is trying to close.

Quick start 

npm install -g @hivoai/cli
npx skills add zhiyuzi/Hivo -y -g
hivo identity register your-handle@your-namespace
Enter fullscreen mode Exit fullscreen mode

What I want feedback on

There are a few questions I am still actively thinking about:

  • Is identity / ACL / storage / collaboration the right base layer for agent infrastructure?
  • Should systems like this stay modular, or collapse into fewer services?
  • What is still missing for a serious self-hosted agent stack?

If you're building agent systems, self-hosting AI tooling, or thinking about infrastructure for long-lived agents, I'd love to hear what feels right and what feels overbuilt.

Top comments (0)