LiteLLM on Render

Deploy LiteLLM, the OpenAI-compatible LLM gateway, on Render with the official Docker image and managed PostgreSQL. Add models in the Admin UI; call the proxy from any OpenAI SDK.

Live Admin UI: litellm-l9il.onrender.com/ui (user admin, password is the master key). Chat playground against that proxy: litellm-tester.onrender.com.

What This Template Shows

This repo wraps BerriAI's official monolithic image as a one-click Render Blueprint. It matches the upstream docker-compose quickstart: one proxy process plus Postgres. Redis is not included, because LiteLLM only requires it when you run more than one instance.

Piece	Role
LiteLLM	OpenAI-compatible gateway, virtual keys, spend logs, Admin UI
litellm/litellm:v1.99.1	Official image with Prisma migrations bundled
Render Web Service	Pulls the image on a Pro instance (4 GB)
Render PostgreSQL	Keys, teams, models, and spend logs

Provider keys (OpenAI, Anthropic, and the rest) are added in the UI after login. Render does not offer an LLM gateway primitive; this template runs LiteLLM's own proxy.

Architecture

flowchart LR
  apps["Apps / OpenAI SDKs"] --> web["litellm<br/>litellm/litellm:v1.99.1"]
  browser["Browser /ui"] --> web
  web --> db[("litellm-db")]
  web -.-> providers["Your LLM providers"]

Loading

How It Works

1. Click Deploy to Render. Render forks this template into your GitHub account and applies render.yaml.
2. On Apply, paste LITELLM_MASTER_KEY and LITELLM_SALT_KEY. Both must start with sk-. OPENAI_API_KEY can be left blank.
3. Render pulls docker.io/litellm/litellm:v1.99.1, wires DATABASE_URL, and starts one Uvicorn worker on port 4000.
4. Open https://<service>.onrender.com/ui. Username is admin. Password is the master key.
5. Add a model, create a virtual key, then call /v1/chat/completions.

Resource	Type	Plan	Notes
litellm	Web (runtime: image)	pro	Health check /health/liveliness. Auto-deploy off.
litellm-db	PostgreSQL 16	basic-256mb	Internal URL only (ipAllowList: []).

Default region: oregon. The proxy is stateless; all durable state lives in Postgres. Horizontal scale needs a Render Key Value instance for Redis, which this Blueprint does not create.

Quick Start

Prerequisites

• A Render account
• Two secrets that start with sk- (master key and salt key)
• An LLM provider key when you are ready to add a model (not required at Apply)

Generate the two LiteLLM secrets:

echo "sk-$(openssl rand -hex 24)"
echo "sk-$(openssl rand -hex 24)"

Keep the salt key. Changing it after you add models makes stored provider credentials unreadable.

Deploy

1. Click Deploy to Render above and fork into your GitHub account.

   

2. On Apply, set LITELLM_MASTER_KEY and LITELLM_SALT_KEY. Leave OPENAI_API_KEY empty unless you want it available as os.environ/OPENAI_API_KEY in the UI.

3. Wait until the web service is Live (~4–10 minutes; first image pull and Prisma migrate can take longer).

4. Open /ui, sign in as admin with the master key, and add a model.

5. Create a virtual key and send a test completion.

curl --fail --silent "https://<your-service>.onrender.com/health/liveliness"

curl "https://<your-service>.onrender.com/v1/chat/completions" \
  -H "Authorization: Bearer <virtual-key>" \
  -H "Content-Type: application/json" \
  -d '{"model":"<your-model>","messages":[{"role":"user","content":"Say hello in five words."}]}'

Features

Feature	Description
Official image	Pinned litellm/litellm:v1.99.1 (also published at ghcr.io/berriai/litellm)
Admin UI	Models, virtual keys, spend, and playground at /ui
OpenAI-compatible API	Point any OpenAI SDK at the service URL
Postgres-backed keys	Virtual keys, budgets, and spend logs survive deploys
UI model catalog	STORE_MODEL_IN_DB=True so you do not need a config file
Single worker	NUM_WORKERS=1 matches LiteLLM's Kubernetes guidance

Configuration

Variable	Source	Description
PORT	Blueprint	4000 (image listen port)
HOST	Blueprint	0.0.0.0
NUM_WORKERS	Blueprint	1
LITELLM_MODE	Blueprint	PRODUCTION (disables load_dotenv)
LITELLM_LOG	Blueprint	ERROR
STORE_MODEL_IN_DB	Blueprint	True: manage models from the UI
DATABASE_URL	Wired	Internal connection string from litellm-db
LITELLM_MASTER_KEY	Required (sync: false)	Admin API key and UI password. Must start with sk-
LITELLM_SALT_KEY	Required (sync: false)	Encrypts provider keys in Postgres. Set once
OPENAI_API_KEY	Optional (sync: false)	Leave blank to add keys in the UI instead

Other provider keys (ANTHROPIC_API_KEY, Azure, Bedrock, and so on) can be added later in the Dashboard or as os.environ/... references in the model form. Full list: LiteLLM docs.

Pin or bump the image

# render.yaml
image:
  url: docker.io/litellm/litellm:v1.99.1

autoDeployTrigger: off so tag edits do not redeploy until you choose Manual Deploy.

Cost

Approximate monthly compute from Render pricing (Hobby workspace, Oregon):

Resource	Approx. monthly
Web service (Pro, 4 GB)	$85
PostgreSQL (Basic 256 MB)	$6
Total	~$91

LLM tokens are billed by your providers. Pro is the floor. LiteLLM documents 1 vCPU and 4 GiB per worker because Prisma's query engine holds a high-water memory mark. Downgrading to Standard (2 GB) typically OOMs and Render reports "No open ports detected."

Spend logs can grow. If Postgres disk fills, raise the database plan. Adding Redis later (for multiple instances) is about $10/month on Key Value Starter.

Troubleshooting

Problem	Solution
Health check fails / no open ports	Keep Pro. Confirm PORT=4000 and HOST=0.0.0.0. First boot runs Prisma migrate; wait and retry.
Reached heap limit / OOMKill	Undersized instance. Do not use Starter or Standard.
Image pull failures	Confirm docker.io/litellm/litellm:v1.99.1 exists, then retry the deploy.
UI login fails	Username is admin. Password is LITELLM_MASTER_KEY and must start with sk-.
Prisma / self-signed certificate	Internal DATABASE_URL should not require TLS. If you pasted an external URL, switch back to the Blueprint fromDatabase value.
Models unreadable after a restart	You rotated LITELLM_SALT_KEY. Restore the original salt; there is no in-place rotation.

Project Structure

render.yaml       Render Blueprint (image + Postgres)
README.md         This file
LICENSE           MIT (template wrapper)
.env.example      Optional env overrides
assets/           Hero / logo

Learn More

Render:

• Web Services
• PostgreSQL
• Blueprints
• Docker images

LiteLLM:

• Proxy quickstart
• Production deployment
• Production checklist
• GitHub

License

MIT for this template wrapper.

Upstream LiteLLM is MIT for the open-source tree. The enterprise/ directory in that repo has a separate license and is not part of this image wrapper.