NanoGPT + LibreChat Setup Guide 2026

librechat is the best open-source chatgpt alternative i've used. NanoGPT gives it access to 400+ models. together, they're basically the ultimate self-hosted AI setup - and it costs me about $8/month total.

tl;dr: LibreChat plus NanoGPT is the ultimate self-hosted AI setup for $8/month. it runs on a Raspberry Pi 5, supports 400+ models, has conversation search, presets, and multi-user sharing. setup takes 15 minutes with Docker.

Key Takeaways:

LibreChat + NanoGPT costs about $8/month total, running on hardware as small as a Raspberry Pi 5
setup takes 15 minutes: clone, configure .env with NanoGPT key, docker compose up, done
LibreChat has more plugins and customization than Open WebUI, with better conversation organization via folders and tags

i run this on a raspberry pi 5 at home. if that tells you anything about how lightweight it is.

what is librechat

librechat is an open-source AI chat interface you host yourself. think of it as a self-hosted chatgpt that supports multiple AI providers. it's free, runs on your own server, and your data stays on your machine.

why use it with NanoGPT specifically:

benefit	explanation
self-hosted	your data stays on your server
multi-model	NanoGPT provides 400+ models
free UI	no subscription for the interface
customizable	themes, plugins, behavior tweaks
privacy	no tracking, no data collection
multi-user	share with family or small team

compared to using NanoGPT's web interface directly, librechat gives you a better UI, conversation history, presets, and plugins. it's worth the 15-minute setup.

what you need

a NanoGPT account with credits and an API key (sign up here)
Docker installed on your machine or server
Git installed
15 minutes (30 if Docker is new to you)

system requirements

short answer: minimum 2GB RAM, 5GB storage, 1 core CPU. runs fine on Raspberry Pi 4/5 with 4GB RAM.

component	minimum	recommended
RAM	2 GB	4 GB
storage	5 GB	10 GB
CPU	1 core	2 cores
OS	linux, mac, windows	linux (ubuntu 22.04+)

runs fine on a raspberry pi 4/5 with 4GB RAM. the heavy lifting happens on NanoGPT's servers - librechat is just the UI.

installation

step 1: clone librechat

short answer: git clone the LibreChat repo and cd into the directory.

git clone https://github.com/danny-avila/LibreChat.git
cd LibreChat

step 2: configure environment

short answer: copy .env.example to .env and add your NanoGPT API key and base URL.

cp .env.example .env

edit .env and add your NanoGPT configuration:

NANOGPT_API_KEY=your-nanogpt-api-key-here
NANOGPT_BASE_URL=https://api.nano-gpt.com/v1

step 3: start with Docker

short answer: docker compose up -d starts LibreChat and MongoDB. first run takes a few minutes.

docker compose up -d

this starts librechat and its MongoDB database. first run takes a few minutes to download images. grab a coffee.

step 4: access librechat

short answer: open http://localhost:3080. create your admin account on first visit.

open your browser, go to:

http://localhost:3080

create your admin account on first visit.

step 5: add NanoGPT as endpoint

short answer: in Settings, Endpoints, Add Custom Endpoint with NanoGPT base URL, API key, and model list.

go to Settings (gear icon)
click Endpoints
click Add Custom Endpoint
fill in:

Name: NanoGPT
Base URL: https://api.nano-gpt.com/v1
API Key: your NanoGPT API key
Models: select the models you want (or add manually)

click Save

step 6: test it

short answer: select a model from the dropdown, send a test message. if you get a response, you're done.

select a model from the dropdown and send a test message. if you get a response, you're done. if not, check the troubleshooting section below.

model configuration

if models don't appear automatically, add them manually:

go to Settings → Endpoints → NanoGPT
in the Models field, add model IDs:

gpt-4o
claude-3-5-sonnet
gemini-1.5-pro
mistral-large
deepseek-v3
gpt-4o-mini

my recommended presets

short answer: gpt-4o for general chat, Claude 3.5 for writing, DeepSeek for budget. create presets with different temperatures for each task.

use case	model	why
general chat	gpt-4o	best all-around
writing	claude 3.5 sonnet	better prose
coding	gpt-4o	most consistent code
quick answers	gpt-4o-mini	fast and cheap
long documents	gemini 1.5 pro	huge context window
budget use	deepseek v3	10x cheaper than gpt-4o

advanced stuff

custom system prompts

short answer: set default system prompts in Settings. keep them focused for better results.

go to Settings → System Prompt and set defaults:

You are a helpful assistant. Be concise and direct.

presets for different tasks

short answer: create presets with different models and temperatures: coding at 0.3, creative writing at 0.9, quick answers at 0.5.

click the Preset dropdown
click Create Preset
configure name, model, system prompt, temperature, max tokens
save

i have presets for "coding helper" (gpt-4o, temp 0.3), "creative writing" (claude sonnet, temp 0.9), and "quick answers" (gpt-4o-mini, temp 0.5).

organizing conversations

short answer: use folders to group by project, full-text search across all conversations, export as JSON or markdown, pin important chats.

librechat supports folders and tags:

folders - group by project or topic
search - full-text search across all conversations
export - download as JSON or markdown
pin - keep important chats at the top

i separate work and personal conversations with folders. simple but effective.

multi-user setup

short answer: set ALLOW_REGISTRATION=true in .env. others create accounts at your server URL. shared NanoGPT API key and balance.

for sharing with family or a small team:

ALLOW_REGISTRATION=true

others create accounts at http://your-server:3080. each user gets their own conversation history. all users share the same NanoGPT API key and balance.

troubleshooting

problem	cause	fix
"invalid API key"	wrong key or base URL	check settings in Endpoints
no models showing	endpoint not configured	add models manually
slow responses	server resources	increase Docker memory limit
connection refused	Docker not running	run `docker compose up -d`
"rate limited"	too many requests	wait or top up NanoGPT balance
blank responses	model error	try a different model

Docker commands you'll need

short answer: docker compose ps to check status, docker compose logs for debugging, docker compose down && up -d to restart.

# check if containers are running
docker compose ps

# view logs
docker compose logs

# restart everything
docker compose down && docker compose up -d

# rebuild after updates
git pull
docker compose build
docker compose up -d

testing your API key directly

short answer: curl the NanoGPT models endpoint. if it returns model data, your key works and the issue is in LibreChat config.

curl https://api.nano-gpt.com/v1/models \
  -H "Authorization: Bearer YOUR_KEY_HERE"

if this returns a list of models, your key works. the issue is in librechat configuration.

performance tuning

short answer: allocate 2GB+ RAM to Docker, use SSD storage, reduce context size, close unused conversations for faster responses.

if librechat feels slow:

allocate at least 2GB RAM to Docker
use SSD storage (MongoDB loves SSDs)
reduce context size in model settings
close unused conversations
use faster models (gpt-4o-mini responds faster than gpt-4o)

librechat vs other interfaces

feature	librechat	open webui	NanoGPT web
self-hosted	yes	yes	no
multi-model	yes	yes	yes
plugins	yes	limited	no
multi-user	yes	yes	no
mobile friendly	yes	yes	yes
setup difficulty	medium	medium	none
conversation search	yes	yes	basic
presets	yes	yes	no

see our Open WebUI setup guide for an alternative. open webui is easier to set up, librechat has more plugins and customization.

updating librechat

cd LibreChat
git pull
docker compose build
docker compose up -d

check for updates weekly or so. the project is actively maintained.

data privacy

everything runs on your machine. the only external connection is to NanoGPT's API when you send a message. librechat doesn't phone home or collect analytics. your conversations are stored in MongoDB on your server.

if you want maximum privacy, run it on a local machine (not a VPS). if you need remote access, use a VPN or reverse proxy with HTTPS - don't expose port 3080 to the public internet.

my setup

raspberry pi 5 at home. connected to NanoGPT via VPN. total cost: ~$8/month for NanoGPT credits, $0 for librechat.

compared to $20/month for chatgpt plus, i get a better interface, more models, and full control over my data. setup took 30 minutes including Docker installation. if you already have Docker, expect 15 minutes.

👉 Get started with NanoGPT

Last updated: July 2026

NanoGPT + Open WebUI Setup - alternative interface
NanoGPT API Key Setup - get your key first
NanoGPT Pricing - how much you'll spend
Best NanoGPT Models - full model list
NanoGPT Privacy Review - data handling details
SillyTavern + NanoGPT Setup - for roleplay users

Disclosure: This article contains affiliate links. If you sign up through our referral link, you get a 5% discount and we earn a small commission. This doesn't affect our reviews - we pay for all services ourselves.