Skip to main content
LibreChat is an open-source, self-hosted AI chat platform that unifies multiple model providers behind a single ChatGPT-style interface. It ships with built-in support for OpenAI-compatible APIs, so you can point it at SambaNova’s /v1 endpoint and use any SambaCloud model alongside OpenAI, Anthropic, Google, and other providers: no proxy required. This guide shows how to connect LibreChat to SambaNova using the built-in Custom Endpoints feature.

​
Prerequisites

  • A SambaCloud account and API key.
  • A running LibreChat instance. Pick one of the installation methods below.

​
Install LibreChat

LibreChat is a self-hosted web app, not a desktop installer: there is no .exe, .dmg, or AppImage to download. Pick one of the installation methods below to get started, then use it in your browser at http://localhost:3080. Docker and npm run it on your own machine on Windows, macOS, or Linux; the Helm chart deploys to a Kubernetes cluster instead. Everything runs in containers. MongoDB, MeiliSearch, RAG API, and Vector DB are all included automatically. Prerequisites: Git, Docker Desktop

​
npm

Run LibreChat directly with Node.js. You manage external services like MongoDB and MeiliSearch yourself. Prerequisites: Node.js v24.16.0, Git, MongoDB

​
Helm Chart

For Kubernetes deployments, follow the LibreChat Helm chart guide directly. The chart requires generated values for CREDS_KEY, CREDS_IV, JWT_SECRET, JWT_REFRESH_SECRET, and MEILI_MASTER_KEY, plus your provider API keys, before the Secret can be created. For full installation details, see the LibreChat quickstart.

​
Setup

These steps assume the Docker Compose install. On the npm install, LibreChat reads librechat.yaml from the project root directly: skip the docker-compose.override.yml override in step 2, and apply changes by stopping and rerunning npm run backend instead of recreating the container.
1

Locate your librechat.yaml

LibreChat reads its endpoint configuration from librechat.yaml in the project root. The repo ships librechat.example.yaml; copy it to librechat.yaml before editing:
2

Add SambaNova as a custom endpoint

Append a SambaNova entry under endpoints.custom. The base URL is https://api.sambanova.ai/v1 and the API key is your SambaCloud key.
librechat.yaml
Set models.fetch: true to pull the live model list from /v1/models on startup. New SambaCloud models appear in the picker without redeploying.
The official docker-compose.yml does not mount librechat.yaml into the API container. If you started from the Docker Compose template, create a docker-compose.override.yml next to it so the file is bind-mounted into the container:
docker-compose.override.yml
Without this override, the API logs Config file YAML format is invalid: ENOENT on startup and the SambaNova endpoint is never registered.
3

Provide the API key

Add the key to your .env file (or the environment passed to the LibreChat API container):
.env
LibreChat resolves ${SAMBANOVA_API_KEY} at startup. The next step applies both this key and the config change.
4

Recreate the API container

The override adds a new bind mount, and docker compose restart does not re-read your Compose configuration. Recreate the container so the mount takes effect:
5

Create your account

Open http://localhost:3080. LibreChat enables registration by default, so the first visit shows a login screen. Click Sign up and create an account.
6

Verify the SambaNova models

Open the model picker and confirm the SambaNova models appear under the SambaNova group. Start a new chat, pick a model, and send a test prompt.
LibreChat interface showing a chat with the prompt "What can you do?" answered by the default SambaNova model

​
Troubleshooting

Verify SAMBANOVA_API_KEY is set in the environment passed to the API container, then run docker compose up -d. A plain docker compose restart will not pick up a newly added bind mount. Check container logs with docker compose logs api for config-load errors.
Regenerate your API key in the SambaCloud console and update SAMBANOVA_API_KEY in .env.
Add customParams.includeReasoningHistory: true to the endpoint block and restart. This flag implies includeReasoningContent and also reconstructs reasoning_content from persisted history across turns.

​
Additional resources