Control Plane for AI Agents

Command Your AI Fleet From Anywhere

NextAgent is how we run AI coding agents for our clients — and how clients stay in control of their projects. Real-time streaming, full tool visibility, integrated email, zero SSH.

Terminal access doesn't scale

Running AI coding agents means SSH sessions into every machine, switching terminals to check on progress, and hoping nothing breaks while you're looking at another screen. There's no shared visibility, no audit trail, and no way for clients to see what their agents are doing.

We built NextAgent because we needed it. Managing dozens of agents across client projects from terminal windows wasn't working — and our clients deserved direct access to their agents, not secondhand status updates.

NextAgent is our single pane of glass. Every agent, every machine, every conversation — in one place. We use it to deliver, and our clients use it to stay in the loop.

What's inside

A complete control plane for AI coding agents — from deployment to monitoring to email.

Real-Time Streaming Chat

Watch your agent think, reason, and act — token by token. Collapsible thinking panels show the full reasoning chain. Tool use cards display live status as files are read, code is written, and tests are run.

Zero-Config Machine Management

A lightweight Python daemon connects outbound over HTTPS — no open ports, no SSH, no VPN. Works behind NAT, corporate firewalls, and cloud VPCs. One command to install, a 6-digit code to pair.

18 Agent Skills

Structured workflows that shape how agents operate: systematic debugging, plan-review gates, TDD, architecture reasoning, code review, branch management, and documentation lookup. Assign per-agent.

Encrypted Service Keys

AES-256-GCM encrypted API keys deployed to agent machines automatically. Agents access search, image generation, and external APIs without credentials ever appearing in the browser.

Integrated Mail

Connect any IMAP/SMTP account. Read, compose, and reply to email from the sidebar. Forward messages to your agent as context. Agents draft replies for your review — they never send autonomously.

Context Engine

Before each session, NextAgent computes git status, cached verification results, and a full file index — injected automatically so agents start with complete situational awareness.

Up and running in minutes

We handle setup for client projects. Here's what it takes.

01

Install

curl -fsSL nextagent.hostkit.dev/install.sh | bash

One command installs the lightweight Python daemon. No Docker, no config files, no dependencies beyond Python 3.11.

02

Pair

Enter a 6-digit code in the browser to link your machine. Register project directories as agent workspaces.

03

Command

Chat with your agents from any browser. They see your code, run your tools, and report back in real time.

How NextAgent connects

The NAEP protocol — optimized JSONL over SSE — streams agent events from remote machines to your browser in real time.

BrowserREST + SSE
You / Your Client
NextAgent ServerPostgres + S3 + Auth
NAEP Protocol
Agent Machinesna-host daemon
Outbound HTTPS

No inbound ports. The na-host daemon initiates an outbound connection — works behind NAT, corporate firewalls, and cloud VPCs. Claude CLI runs as a native subprocess on the machine, with full access to local tools and project files.

Supervised Mode

Review Before Execute

In supervised mode, agents write implementation plans before touching code. You see every proposed change, approve or redirect, and maintain human oversight at every step. Agents can also run autonomously or in research-only mode — configurable per agent.

Plan awaiting review

## Implementation Plan

1. Add middleware to validate JWT tokens

2. Create protected route wrapper component

3. Update API endpoints with auth checks

Approve
Request Changes

Re: API integration timeline

Quick question about the auth endpoint...

2m ago

Deploy confirmation — staging

Build succeeded, all checks passed...

1h ago
Forward to Agent
Reply
Built-In Mail

Email Meets Agent

Connect any IMAP/SMTP account — Gmail, Outlook, self-hosted, anything. Your inbox lives in the sidebar alongside your agents. Forward a message to an agent and it becomes structured context for the conversation.

Agents can draft replies for your review, but they never send autonomously. You see a preview card with a Send button — the human always has the final say.

Rich Rendering

More Than Text

Mermaid diagrams render inline with a dark theme. HTML blocks render in Shadow DOM with full navigation. Code blocks get syntax highlighting. File paths in chat are clickable links that open directly in a tab panel.

Voice input via push-to-talk. Image generation with automatic S3 publishing. Your agent communicates in the richest format the task demands.

Agent output

mermaid

Client
API
DB

typescript

export async function getUser(id: string)

Put your agents to work

NextAgent is how Emergent delivers — and how our clients stay in control. Talk to us about setting up agents for your projects.

Emergent

What are you working on?

Powered by Claude · Responses may vary