MultiAgent puts agent factories on your servers. A factory — the place where agents are assembled and run: it has its own deployment slug, folders, a shared container network, a database, a Git server, Nginx routing, a home page and a login. A server — the physical machine a factory lives on; one machine can host several factories. MultiAgent connects to any number of Linux servers over SSH, deploys agents and products on their factories (Research, Analytics, Data, Proxy, MultiAgent itself, plus chatbots, Telegram, Testing, News, Portal and CRM), collects their chats and analytics into one interface, and hands out app access through the factory’s subfolders via Nginx routing.
System requirements: MultiAgent itself — Linux with Docker, or Windows 10 and newer; 1 GB of memory is enough. The servers it manages are Linux (Ubuntu/Debian) with SSH access and root/sudo rights. MultiAgent installs Docker and everything else on the factories by itself.
Installation
MultiAgent is installed on one machine — your own server or a Windows computer: in Docker it is a single container with a persistent volume for data, on Windows an installer that registers a service. The agents and apps do not run here but on the managed servers: MultiAgent is only the console.
The step-by-step route — from requirements and installation to the first working factory — is on the Installation & Setup page. The details of each method and the environment settings: Installation in Docker and Installation on Windows.
First Login
MultiAgent opens at the address where it is deployed (locally — http://localhost:3990). The first user is created with the “Register” button on the login page and remains the only one — that is the administrator. The login and password can also be set in advance — with the MULTIAGENT_LOGIN and MULTIAGENT_PASSWORD variables. Authentication uses JWT tokens valid for 90 days, so you will rarely need to sign in again.
Main Screen
After signing in you land on the main screen headed “Manage factories and fleets of agents”, with shortcut tiles for the product’s sections: servers, factories, fleets, agents, chats, tokens, cost, speed, hardware, incidents, licenses and backup — each with a one-line hint. It is the same map the sidebar carries, only with the hints written out: nothing to memorize. The permanent navigation tree lives in the left sidebar. There are three top-level sections: Factories, Management and Assets. Your factories live under “Factories”, and under each factory its agents and apps: nothing needs adding to the tree by hand, everything appears on its own. A right-click on a row (or the three-dot button) opens the action menu, and its first item is “Open”: for a factory it is its home page, for an agent or app its public address. The menu selects the row first, so its page opens right after the menu. The status next to the row updates right there, without leaving the page.
There is no flat page listing all agents: agents are visible under their factory’s row, and the fleet’s full list is given by “Management → Agents”. It is convenient to watch the chats of all agents in “Management → Chats”.
- Factories — the place where agents are assembled: installing Docker on the machine, folders and databases, Nginx routing and certificates, the home page, central login, backups and handing a factory to the client. Details — Factories;
- Management — the control room for a fleet, once there are many agents. It holds eight screens in work order: “Tokens”, “Cost”, “Speed”, “Chats”, “Hardware”, “Agents”, “Decisions” and “Incidents”. The root row of the section itself opens the fleet overview. Details — Management;
- Assets — a folder without a page of its own: clicking it unfolds the list but takes you nowhere. Inside are “Servers”, “Fleets”, “Policies”, “Licenses”, “Models”, “Skills”, “Environments”, “Files”, “Tools” and “Accounts”.
From “Assets” it is convenient to hand the same settings out to all agents at once: “Models” and “Skills” are catalogs whose configuration or set of files goes to the agents you pick (the agent’s record with the same name is replaced), “Environments” are named sets of environment variables, “Servers” are the machines’ records with their connections, and “Fleets” and “Policies” are where agent groups and their rules are configured. “Files” is for viewing and editing factory files together with a terminal, “Tools” is a library of Docker service templates, “Accounts” is for LLM accounts with quotas and balances, and “Licenses” is the agents’ license keys.
When the AGENT_API_URL and AGENT_API_KEY environment variables are set, an Own Agent panel appears at the bottom of the sidebar: a list of active chats with timers and token and cost totals, plus the “Tasks (N)” button — the queue of long jobs with a live counter — and “Chats”. The panel opens from any page and switches between chat and tasks in place. Details in Own Agent and Tasks.
The Typical Path
- Connect a server — the machine the factories will live on: address, SSH port, user, password or private key. When MultiAgent runs on Windows, choose the Local WSL mode instead — a distribution on this very computer, with no password or key. Details in Servers.
- Add a factory — pick the server, set the name and the deployment slug (default
agents). The slug is the short name that separates this factory from the others on the same machine: it sets the network, the container-name prefix and the folders. Details in Factories. - Run “Check and fix” — the button on the factory page opens a checklist: the connection, Docker, folders, the agent user, Nginx, the certificate, Postgres, Forgejo, backups, the home page, the central login. The rows missing something are repaired in one click: Docker and the rest MultiAgent installs itself.
- Add an agent — the “New Agent” wizard: pick the type — a standalone agent or a product (Research, Analytics, Data, Proxy, MultiAgent, Chatbot, Telegram, Test, News, Portal, CRM); the name, factory, image and port are filled in for you. Details in Agents.
- Deploy the agent or product — MultiAgent generates compose.yml and brings the container up on the machine. Products deploy together with a linked agent: by default the factory’s first standalone agent is used, and if there is none, a new one is created by the same wizard.
- Enable the route — install Nginx from the factory’s tab, and the agent immediately gets its own subfolder
{host}/{slug}/on the factory. Details in Nginx Routing. - Set up the home page — MultiAgent builds a static card board at the factory’s root: agents, apps and tools, while the “QR login” button in the corner opens sign-in from a phone. No separate container and no separate login — Nginx is enough. Details in Factory Home Page.
- Turn on central login — the “Authorization” (Authelia) tab on the factory page: a visitor signs in once and gets access to every protected app on the factory. It needs HTTPS — a local certificate from the Nginx tab, or the factory’s public address (an address template like
https://{host}/{slug}works too). Details in Factories. - Open the chats — chats from all agents appear in one shared list. Details in Chats.
Interface Language and Time Zone
The interface is fully translated into Russian. The language (English/Russian) and the time zone are chosen in the “Appearance” dialog. The language is saved separately for each user. The time zone is one for the whole system: every operator sees it, and every date, chart and report is shown in it. Time is still stored as UTC, so changing the zone recalculates nothing — only how dates are displayed on screen.