Skip to content

Overview

Nythronix is a control plane for remote machines. A small agent installed on each machine connects to the control server over an outbound WebSocket connection. You can then work with the machine from your browser: open terminals, relay SSH through the control plane, grant one-time access by code, and connect AI workers that run commands with your approval.

The key property: machines need no public IP, open ports, or VPN. The agent behind NAT calls out to the server; the server never initiates a connection to the machine.

Nythronix comes in two forms: SaaS (the dashboard at https://app.nx.3fe.ru, with agents connecting through nx.3fe.ru) and self-hosted (the same codebase on your server). This documentation describes the SaaS dashboard; self-hosted differences are covered in the FAQ.

Component Purpose Runs on
Control plane HTTP/WebSocket API, authentication, auditing, byte routing between agents, chat, and worker memory. nx.3fe.ru
Dashboard Web interface: overview, terminals, tunnels, workers, and administration. https://app.nx.3fe.ru
Agent nythronix-agent A service on the managed machine. Creates terminals (PTY), accepts relay streams, runs worker commands, and updates itself. Each machine
Worker An AI channel participant. Reads the channel, reasons, calls tools, and asks for human approval. A container on the Nythronix side
Operator CLI nyx-operator Lists the fleet and opens a local port routed through the control plane to a machine behind NAT (SSH, scp, rsync). Your laptop
MCP server nyx-mcp Exposes the same actions to Claude Code and other MCP clients: find a machine, open a tunnel, or run a read-only check. Your laptop

See SSH relay and tunnels for a detailed connection diagram.

  • Machine / agent. In the dashboard these mean the same thing: a record of an installed agent, with a login, hostname, version, platform, presence (Online / Offline), and tags.
  • Organization. The isolation boundary. Operators see only their organization’s machines; relays and tunnels work only between agents in the same organization.
  • Operator. A person’s dashboard account. Roles: owner, admin, operator, viewer. Viewers can watch terminals but cannot type in them. Admins and owners can access the Admin section (agent enrollment, tokens, releases, and audit).
  • Session. One terminal on a machine: a plain shell or a session inside tmux or zellij. Sessions live on the agent, not in the browser, and can be rejoined by one or several people. See Terminal and sessions.
  • Enrollment token (nyxet_…). A single-use or reusable token that lets an agent obtain permanent credentials during installation. Create one in Admin → Agents → Enrollment tokens.
  • Relay. A mechanism for switching bytes between two agents, or between your nyx-operator and an agent. This makes SSH and other TCP services reachable behind NAT. Enable it separately on each target machine.
  • Tunnel. A local TCP port that leads through the relay to another machine’s loopback port. It can run on an agent (created in the dashboard) or on your laptop (nyx-operator tunnel).
  • Consent code. A one-time code generated by the machine owner with nythronix-agent share. The operator enters it on the dashboard and receives exactly the access the owner allowed. See Access by consent code.
  • Channel. A chat within an organization for people and workers. A worker can belong to several channels, and a channel can contain several workers.
  • Worker. An AI channel participant. Runs commands on machines only through the control server and with human approval, except for commands on the safe read-only list. On Linux, its commands run as the unprivileged nyx-worker user.
  • Skill. An additional set of worker tools: an MCP server running on the machine itself, for example to access a local database, and made available through the control plane. Enable skills in Admin → Agents → Skills.
  • Administrators managing machines behind NAT (offices, branches, home PCs, VPS instances) who want browser access without a VPN.
  • Teams that need shared terminal viewing and an activity log.
  • People providing one-time support on someone else’s machine who want access to last only as long as the owner permits.
  • Anyone who wants AI workers to handle routine tasks while retaining approval of every command that makes changes.

Start with the quickstart.