Run AnythingLLM on Bare Metal Without Docker

Running AnythingLLM on bare metal means installing and launching it directly on your operating system without any container runtime. This approach gives you full control over the process environment, file permissions, and system integration — and is useful when Docker is unavailable or unwanted. The setup involves cloning the repository, building the React frontend, configuring the server environment, running Prisma database migrations, and starting two Node.js processes: the main server and the document collector.

This deployment method is not officially supported by the AnythingLLM core team. It is provided as a community reference. You are fully responsible for securing your deployment, managing updates, and resolving any issues that arise. Questions specific to bare-metal deployments will not be answered by the core team. For a supported path, use the Docker installation instead.

Prerequisites

Requirement	Minimum Version	Notes
Node.js	v18 or higher	Required by the `engines` field in `package.json`
Yarn	Any recent version	Used for dependency installation and workspace scripts
Git	Any	For cloning the repository and pulling updates
Disk space	10 GB minimum	More if storing large documents or local models
RAM	2 GB minimum	4 GB or more recommended

Setup

Clone the repository

Clone the repo into the directory where you want AnythingLLM to live, as the OS user that will run the application:

git clone git@github.com:Mintplex-Labs/anything-llm.git
cd anything-llm

Install all dependencies

Run the setup script from the repository root. This installs dependencies for all three workspaces (server, collector, and frontend) in a single step:

yarn setup

The setup script also copies the example .env files into place for development use and runs the initial Prisma database generation and migration. After it completes, you will have:

server/.env.development (copied from server/.env.example) — note: this is the development env file; Step 3 creates the production one
collector/.env (copied from collector/.env.example)
docker/.env (copied from docker/.env.example)
frontend/.env (copied from frontend/.env.example)

Configure the server environment

Copy the server environment file for production use:

cp server/.env.example server/.env

Open server/.env in your editor and set at minimum the STORAGE_DIR variable to an absolute path where AnythingLLM will store its database, uploaded documents, and vector indexes:

STORAGE_DIR="/your/absolute/path/to/server/storage"

This file will be read on every server start and can be updated while the service is stopped to change settings. The full list of available environment variables is documented in docker/.env.example.

Configure the frontend for production

Edit frontend/.env and set VITE_API_BASE to /api. This tells the compiled frontend to call the API relative to wherever it is served from — required for production deployments:

# frontend/.env
VITE_API_BASE='/api'

Do not use the http://localhost:3001/api value here — that is only for local development.

Build the frontend

Compile the React/Vite frontend into static files:

cd frontend && yarn build

This produces a frontend/dist directory containing the compiled HTML, CSS, and JavaScript. The build step only needs to be re-run when you update AnythingLLM or make frontend changes.

Copy the frontend build to the server

Copy the compiled frontend assets into the server’s public directory, which is where the Node.js server will serve them from:

cp -R frontend/dist server/public

After this step, server/public/index.html should exist. The server will serve the React app from this location.

Set up the database

Run the Prisma migrations to create and initialise the SQLite database:

cd server && npx prisma generate --schema=./prisma/schema.prisma
cd server && npx prisma migrate deploy --schema=./prisma/schema.prisma

The first command generates the Prisma client. The second applies all migrations to create the database schema. Both commands must be run from the server directory, and should be re-run after every update that includes new migrations.

Start the server and collector

AnythingLLM requires two separate processes running concurrently. Start each in its own terminal session (or as background processes):

# Terminal 1 — main API server
cd server && NODE_ENV=production node index.js &

# Terminal 2 — document collector
cd collector && NODE_ENV=production node index.js &

Both processes must be running for the full application to work. The server handles API requests, LLM calls, and the web UI. The collector handles document parsing and ingestion.AnythingLLM is now available at http://localhost:3001.

Updating AnythingLLM

To update to the latest version, pull the new code and repeat the build and migration steps:

# Stop running processes first
pkill node

# Pull latest changes
git pull origin master

# Reinstall dependencies in case packages changed
cd $HOME/anything-llm/collector && yarn
cd $HOME/anything-llm/server && yarn

# Rebuild the frontend
cd $HOME/anything-llm/frontend && yarn && yarn build && cd $HOME/anything-llm

# Copy new frontend build
rm -rf server/public
cp -r frontend/dist server/public

# Run any new database migrations
cd $HOME/anything-llm/server && npx prisma migrate deploy --schema=./prisma/schema.prisma
cd $HOME/anything-llm/server && npx prisma generate

# Restart both processes
cd $HOME/anything-llm/server
(NODE_ENV=production node index.js) &> /logs/server.log &

cd $HOME/anything-llm/collector
(NODE_ENV=production node index.js) &> /logs/collector.log &

Always run pkill node before updating to avoid running multiple AnythingLLM processes simultaneously, which can cause database conflicts and unpredictable behavior.

Example Update Script

The following shell script automates the full update procedure. Save it as update-anythingllm.sh and run it whenever you want to update:

#!/bin/bash

cd $HOME/anything-llm &&\
git checkout . &&\
git pull origin master &&\
echo "HEAD pulled to commit $(git log -1 --pretty=format:"%h" | tail -n 1)"

echo "Freezing current ENVs"
curl -I "http://localhost:3001/api/env-dump" | head -n 1|cut -d$' ' -f2

echo "Rebuilding Frontend"
cd $HOME/anything-llm/frontend && yarn && yarn build && cd $HOME/anything-llm

echo "Copying to Server Public"
rm -rf server/public
cp -r frontend/dist server/public

echo "Killing node processes"
pkill node

echo "Installing collector dependencies"
cd $HOME/anything-llm/collector && yarn

echo "Installing server dependencies & running migrations"
cd $HOME/anything-llm/server && yarn
cd $HOME/anything-llm/server && npx prisma migrate deploy --schema=./prisma/schema.prisma
cd $HOME/anything-llm/server && npx prisma generate

echo "Booting up services."
truncate -s 0 /logs/server.log
truncate -s 0 /logs/collector.log

cd $HOME/anything-llm/server
(NODE_ENV=production node index.js) &> /logs/server.log &

cd $HOME/anything-llm/collector
(NODE_ENV=production node index.js) &> /logs/collector.log &

Using Nginx as a Reverse Proxy

If you want to serve AnythingLLM on port 80 or behind a domain name, use Nginx as a reverse proxy. Streaming chat responses and agent protocol both require WebSocket support — the configuration below handles both:

server {
   # Enable WebSocket connections for agent protocol.
   location ~* ^/api/agent-invocation/(.*) {
      proxy_pass http://0.0.0.0:3001;
      proxy_http_version 1.1;
      proxy_set_header Upgrade $http_upgrade;
      proxy_set_header Connection "Upgrade";
   }

   listen 80;
   server_name your-domain.example.com;

   location / {
      # Prevent timeouts on long-running LLM requests.
      proxy_connect_timeout       605;
      proxy_send_timeout          605;
      proxy_read_timeout          605;
      send_timeout                605;
      keepalive_timeout           605;

      # Enable readable HTTP streaming for LLM streamed responses.
      proxy_buffering off;
      proxy_cache off;

      proxy_pass http://0.0.0.0:3001;
   }
}

Replace your-domain.example.com with your actual domain. If you plan to use HTTPS, configure SSL termination in Nginx and set ENABLE_HTTPS in the server .env if you want the Node server itself to handle TLS instead.

Running as a System Service

For production deployments, you should manage both processes with a process manager like systemd or pm2 rather than running them in background shell sessions.

Using pm2

# Install pm2 globally
npm install -g pm2

# Start both services
pm2 start "cd /path/to/anything-llm/server && NODE_ENV=production node index.js" --name anythingllm-server
pm2 start "cd /path/to/anything-llm/collector && NODE_ENV=production node index.js" --name anythingllm-collector

# Save the process list so they restart on reboot
pm2 save
pm2 startup

Get Started

Configuration

Core Features

AI Agents

Advanced

Run AnythingLLM on Bare Metal Without Docker

Prerequisites

Setup

Updating AnythingLLM

Example Update Script

Using Nginx as a Reverse Proxy

Running as a System Service

Using pm2

Build docs developers (and LLMs) love

Get Started

Configuration

Core Features

AI Agents

Advanced

Documentation Index

​Prerequisites

​Setup

​Updating AnythingLLM

​Example Update Script

​Using Nginx as a Reverse Proxy

​Running as a System Service

​Using pm2

Build docs developers (and LLMs) love

Prerequisites

Setup

Updating AnythingLLM

Example Update Script

Using Nginx as a Reverse Proxy

Running as a System Service

Using pm2