Self-host LiteLLM
updated Oct 2026deployed by usprices checked · Jul 2026Install verified on Ubuntu 26.04 · Oct 2026 · how we test
We earn commissions when you shop through the links below. Full disclosure →
An open-source LLM gateway: one OpenAI-compatible API in front of 100+ providers (OpenAI, Anthropic, Bedrock, Azure, Ollama and more), with virtual API keys, per-key budgets and spend tracking, routing and fallbacks, and an admin UI. It is a proxy for your apps and tools, not a chat interface.
Key facts
CategoryAI Chat Interfaces
LicenseMIT core with proprietary enterprise modules
StackPython, FastAPI, PostgreSQL
Min RAM1024 MB
Official imageyes
Measured idle RAM610 MB
Difficulty
Our recommendation
Pick LiteLLM when several apps, teammates or tools need to call LLMs and you would rather have one OpenAI-compatible endpoint with your own keys, budgets and spend tracking than scatter provider keys everywhere. It is a gateway, not a chat interface — put it behind Open WebUI or any OpenAI-compatible client — so it is the right pick for the plumbing, and the wrong one if all you want is a chat window.
What you need
- Any VPS with at least 1024 MB of RAM
- A domain you control — most self-hosted setups need HTTPS in front of them
- Under an hour if you've used Docker before
Where to host itaffiliate disclosure
Hetzner Cloudrun it on
From $23.59/mo · 2 vCPU / 4 GB / 80 GB · EU + USKamaterafree trial
From $4/mo · 1 vCPU / 1 GB / 20 GB · US + EU + AsiaDigitalOceanalso works on
From $6/mo · 1 vCPU / 1 GB / 25 GB · US + EU + AsiaPaid link — we earn a commission if you shop through it.
Install
Run these commands on your server:
# LiteLLM — upstream quickstart compose (gateway + Postgres), pinned to release v1.105.0 instead of main-stable
curl -sSLo docker-compose.yml https://raw.githubusercontent.com/BerriAI/litellm/v1.105.0/docker/docker-compose.quickstart.yml
sed -i 's|litellm:main-stable|litellm:v1.105.0|' docker-compose.yml
# master key = root credential and the admin UI password; salt key encrypts provider keys stored in the DB (never regenerate it)
printf 'LITELLM_MASTER_KEY=sk-%s\nLITELLM_SALT_KEY=sk-%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d # → http://SERVER_IP:4000/ui (user admin, password = LITELLM_MASTER_KEY from .env)
# port 4000 is published on all interfaces, no TLS; put a reverse proxy in front. Postgres is not published.What you take on
LiteLLM is quick to start, but it sits in front of every model call you make, so its defaults are worth reading:
non-negotiableLITELLM_MASTER_KEY is the root credential: it authorises every management call and is the admin UI password (user admin). Treat it like a root password and keep .env out of source control.
non-negotiableLITELLM_SALT_KEY encrypts the provider API keys you add in the UI and has no in-place rotation — regenerating it makes stored credentials unreadable. Back up .env along with the Postgres volume.
non-negotiableThe quickstart publishes port 4000 on every interface with no TLS, and its Postgres uses fixed credentials (not published outside the compose network). Put a reverse proxy in front before the gateway holds real provider keys.
non-negotiableThe code is MIT except the enterprise/ directory, which has its own licence. Check that licence before relying on any feature the docs mark as enterprise.
non-negotiableThe project ships fast, with frequent releases. Pin a release tag as the install here does — the main-stable tag the upstream compose uses moves — and read the release notes before upgrading.
More in AI Chat Interfaces
AI Chat Interfaces
AnythingLLM
JavaScriptNode.jsReact
easyRead guide →
AI Chat Interfaces
Jan
Go
moderateRead guide →
AI Chat Interfaces
LibreChat
TypeScriptReactNode.jsMongoDB
moderateRead guide →
AI Chat Interfaces
LobeHub
TypeScriptNext.jsPostgreSQL
easyRead guide →