OpenMake

Self-host / 30-minute path

Install OpenMake with an explicit model boundary.

Choose a hosted OpenAI-compatible endpoint when you need to validate the application without operating inference. Choose a local vLLM or LiteLLM gateway when the model backend must remain under your control.

Download DesktopLive Demo

0. Choose your inference path

Both paths run the same OpenMake application. The difference is where model inference happens and who operates it.

  • Hosted endpoint: faster evaluation and no GPU operation
  • Self-hosted endpoint: operate vLLM or LiteLLM and define your own network, credentials, and model catalog

1. Prepare the runtime

Use Node 24, Docker, PostgreSQL, and an OpenAI-compatible endpoint. Keep secrets outside the repository and set production URLs before public access.

git clone https://github.com/openmake/openmake_llm.git
cd openmake_llm
node --version
docker --version

2. Configure the gateway

Set the default model and the endpoint OpenMake may call. Structural identifiers, model IDs, and environment variable names stay unchanged across locales.

LLM_DEFAULT_MODEL=qwen3.6-35b-a3b
LLM_POOL_DEFAULT_CTX=262144
DB_AUTO_MIGRATE=true

3. Start and verify

Server startup applies pending migrations by default. Visit the web UI, sign in, send one message through the configured model, and confirm the response path before inviting users.

  • Confirm API and web processes are healthy
  • Confirm the configured model appears in the chooser
  • Send a first message and check the audit or usage view

4. Operate deliberately

Use DB_AUTO_MIGRATE=false only when a manual migration process is required. Keep external provider keys optional and separate from the local default route.