Otari on Render

Deploy Otari, Mozilla AI's OpenAI-compatible LLM gateway, on Render with the official mzdotai/otari image and managed Postgres for keys, budgets, and usage.

What This Template Shows

This repo packages Mozilla AI's official Otari image as a one-click Render Blueprint. No Python/uv monorepo build on Render.

Piece	Role
Otari	OpenAI-compatible gateway: 40+ providers, virtual keys, budgets, usage
mzdotai/otari	Official gateway image
Render Web Service	Runs the image (runtime: image)
Render Postgres	Durable keys, budgets, usage, pricing rows

Bring at least one provider key. Point any OpenAI client at your *.onrender.com URL.

Architecture

flowchart LR
  clients["Apps / OpenAI SDKs"] --> otari["otari<br/>mzdotai/otari"]
  otari --> db[("otari-db<br/>Postgres")]
  otari -.->|"provider keys"| providers["OpenAI / Anthropic / …"]

Loading

How It Works

1. Click Deploy to Render. Render forks this template into your GitHub account and applies render.yaml.
2. On Apply, enter at least one provider key (OPENAI_API_KEY, ANTHROPIC_API_KEY, MISTRAL_API_KEY, or GEMINI_API_KEY). Leave unused keys blank.
3. Render pulls mzdotai/otari:0.2.0, starts on port 8000, and wires OTARI_DATABASE_URL from Postgres.
4. On first boot, auto_migrate + bootstrap_api_key create the schema and print a one-time gw-… API key in the logs.
5. Call /v1/chat/completions (or other OpenAI-compatible routes) with that key.

Resource	Type	Plan	Notes
otari	Web (runtime: image)	starter	Health check /health; pinned 0.2.0
otari-db	Postgres 16	basic-256mb	Private (ipAllowList: [])

Default region: oregon. Keep both resources in the same region. Previews are off. This Blueprint does not include optional tool backends (code-exec, web search, guardrails).

Quick Start

Prerequisites

• A Render account
• At least one LLM provider API key

Deploy

1. Click Deploy to Render above and fork into your GitHub account.
2. On Apply, set at least one provider key. Confirm otari (Starter) and otari-db (Basic-256mb).
3. Wait until both are Live (~3–6 minutes; first image pull can take longer).
4. Open the otari service Logs, copy the bootstrap gw-… key printed once on first startup.
5. Hit health, then call the gateway with your OpenAI client pointed at the service URL.

Health check:

curl -sS https://<your-otari-service>.onrender.com/health

Smoke chat (replace URL, key, and model):

curl -sS https://<your-otari-service>.onrender.com/v1/chat/completions \
  -H "Authorization: Bearer gw-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai:gpt-4o-mini","messages":[{"role":"user","content":"hi"}]}'

Features

Feature	Description
Official image	Uses maintained mzdotai/otari tags; no source build
Managed Postgres	OTARI_DATABASE_URL wired via fromDatabase
One-click Blueprint	Web + Postgres via projects / environments
Bootstrap API key	First-boot gw-… key in logs (OTARI_BOOTSTRAP_API_KEY)
Auto migrate	Schema ready on startup (OTARI_AUTO_MIGRATE)
Pinned by default	Image set to 0.2.0; bump deliberately

Configuration

Variable	Source	Description
PORT / OTARI_PORT	Wired	8000 (Otari listen port; both set for Render's scanner)
OTARI_HOST	Wired	0.0.0.0
OTARI_DATABASE_URL	Wired	From otari-db connection string
OTARI_MASTER_KEY	Auto-generated	Management / admin secret
OTARI_REQUIRE_PRICING	Wired	false for env-only deploys without a pricing table
OTARI_AUTO_MIGRATE	Wired	true
OTARI_BOOTSTRAP_API_KEY	Wired	true (prints first-use gw-… key once)
OPENAI_API_KEY	Optional (sync: false)	OpenAI provider
ANTHROPIC_API_KEY	Optional (sync: false)	Anthropic provider
MISTRAL_API_KEY	Optional (sync: false)	Mistral provider
GEMINI_API_KEY	Optional (sync: false)	Gemini provider

See .env.example and upstream configuration docs.

Pin or float the image

# render.yaml
image:
  url: docker.io/mzdotai/otari:0.2.0
  # floating:
  # url: docker.io/mzdotai/otari:latest

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

Cost

Resource	Approx. monthly
Web service (Starter)	~$7
Postgres (basic-256mb)	~$6
Total	~$13

Provider tokens are billed by your LLM vendors. Starter is usually enough for light gateway traffic; raise the web plan if you see OOM or sustained CPU pressure.

Troubleshooting

Problem	Solution
Health check fails / no open ports	Confirm PORT and OTARI_PORT are both 8000. Check logs for migrate errors.
No bootstrap gw-… key	Look at first deploy logs only; key is printed once. Or create a key via management APIs with OTARI_MASTER_KEY.
Provider errors / empty model	Set at least one provider key in Dashboard → Environment.
Pricing / require_pricing failures	Blueprint sets OTARI_REQUIRE_PRICING=false. Add pricing via API or OTARI_CONFIG_YAML when ready.
Image pull failures	Confirm mzdotai/otari:0.2.0 on Docker Hub. Retry deploy.

Project Structure

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

Learn More

Render:

• Web Services
• Deploy an image
• PostgreSQL
• Blueprints
• Deploy to Render button

Otari:

• Upstream repo
• Documentation
• Docker Hub
• Configuration

License

MIT for this template wrapper.

Upstream Otari is Apache-2.0. Star that repo if this helped.