Private AI agent platformself-hosted

AI agents for your whole organization, running entirely on infrastructure you own.

A power user builds an agent once. An admin assigns it. Everyone else just uses it — through chat, on machines inside your own network.

◆ No NextAgent cloud. Nothing of yours to leak, lose, or hand over.
The problem

AI fluency is unevenly distributed.

<1%
of knowledge workers can build and run their own AI agents today.

Every organization has a few technical people who can build genuinely powerful agents with Claude Code. They have no way to put those agents in front of everyone else — the people in finance, ops, legal, support who'd actually use them.

NextAgent is the delivery layer. Build once, assign, and the whole team uses it through plain chat — no terminal, no API keys, no new account on someone else's cloud.

How it works

Three roles. One mental model.

01Admin

Sets the stage

Stands up NextAgent on a machine the org controls, adds people, and decides who can use which agent.

  • Installs on your own VPS, in your VPN
  • Adds users, scopes permissions
  • Assigns agents to teams
02Power user

Builds the agent

Uses the full power of Claude Code on a workspace machine to build an agent once — tools, skills, the works.

  • Full Claude Code, no proprietary framework
  • Bash, Read/Write, MCP servers, skills
  • Ship it to the whole team
03Everyone else

Just uses it

Opens a familiar chat window and talks to the agent. No terminal, no keys, no idea there's a VPS behind it.

  • Plain chat interface
  • Knows who's speaking, scoped per user
  • Zero setup
What you can build

Real agents, for real work.

Anything Claude Code can do, your team can use through chat.

Contract redline

Reads an incoming contract against your playbook, flags risky clauses, and proposes redlines — in the legal team's chat.

Email triage

Sorts the shared inbox, drafts replies in your voice, and routes the hard ones to a human — connected through MCP.

Internal knowledge bot

Answers questions from your own docs and systems — without those docs ever leaving the network they live in.

The trust model

There is no NextAgent cloud.

Privacy through architecture, not policy. The app, your accounts, and every message run on a machine you control. We're not in the path — so there's nothing of yours for us to read, leak, or be compelled to hand over.

Your control plane

Login, users, and all conversation data run on your VPS, at a hostname you choose, inside your VPN.

Your workspace

Agents run on workspace machines with your own Anthropic subscription — compartmentalized to contain blast radius.

Direct inference

Model calls go straight from your workspace to Anthropic. No NextAgent server sits in the middle.

Nothing to subpoena

We don't hold your data, accounts, or messages — so there's nothing for a third party to demand from us.

If NextAgent were breached, the attacker would find no customer data, accounts, or conversations — we don't have them. That's not a promise; it's how the system is built.

Plane 1 · control

The app

Login, users, every message and conversation. The whole control plane.

runs in your network
Plane 2 · execution

Workspace machines

Where agents run — Claude Code projects + your Anthropic subscription. Compartmentalized from the control plane.

runs in your network
Plane 3 · inference

The model

Claude (or a ZDR endpoint) answers the call — reached directly, with no NextAgent hop.

outbound only
Inference

Your models, your terms.

Two paths, both running from hardware you own — no vendor lock-in baked into the platform.

Anthropic · default

Claude, via your own subscription

The harness is Claude Code, running on your workspace machines. It talks directly to Anthropic on your account — and updates itself between you and Anthropic, with NextAgent nowhere in the loop.

OpenRouter · ZDR

Zero-data-retention endpoints

Prefer an OpenAI-style model? An alternative harness routes to ZDR endpoints through OpenRouter — also on your workspace machines, under an admin policy you set.

Pricing

Simple, because you host it.

Self-host
Free
Run it on your own machine. The whole platform, in your network.
Get started
Pro soon
Managed setup, priority support, and team management niceties on top of your own infra.
Request access
Enterprise
Talk to us
Hands-on deployment into your VPN, custom agents, and a same-day support agent of your own.
Contact

Pricing carried over from the current site — confirm before publishing.

FAQ

Questions a careful evaluator asks.

How is this different from ChatGPT Enterprise?

Their model: trust our cloud, we promise not to look. Ours: there's nothing to trust, because we're not in the path. The app and your data live on hardware you own.

What happens to my agents if NextAgent the company disappears?

Nothing. Everything runs on your machines today. There's no service to switch off — the software keeps working.

Do I have to take your word for the privacy claims?

No. The architecture is auditable, and the harness is Claude Code talking directly to Anthropic from your network. You can watch the traffic.

What about compliance?

Because data never leaves your network, NextAgent doesn't add a new processor to your compliance surface. You own the boundary.

Can I use my existing Claude subscription?

Yes — that's the default. Agents run on your Anthropic account from your workspace machines.