LibreChat custom endpoint setup

• LibreChat running from its Docker Compose files, with librechat.yaml in the project root.
• A RelaySwitch key. Sign in with an email code at the console and create one. Keys start with sk-relay-.
• A balance. Top-ups start at $10, by card or crypto, with no subscription. Before the first top-up the console chat runs 12 free turns a day for comparing models.

Put the key in .env and reference it from the YAML. LibreChat's docs call the environment-variable reference the recommended form for apiKey.

• name: the title in the endpoint selector. Required, unique.
• apiKey: ${RELAYSWITCH_KEY} reads the variable from .env. A literal key or user_provided (each user pastes their own) also work.
• baseURL: https://api.relayswitch.io/v1, with the /v1. LibreChat appends /chat/completions itself, as its OpenAI-compatible examples (Mistral at https://api.mistral.ai/v1) show. Leave directEndpoint unset; that is for a URL that is already the full completions path.
• models.default: the ids in the picker. Required; with fetch: false this list is the picker.
• titleConvo and titleModel: the automatic chat title and the model that writes it. Omitted, titles are written by the conversation's model.
• modelDisplayLabel: the label beside the AI icon in the chat; defaults to “AI”.

dropParams is not needed: the API accepts the sampling fields LibreChat sends. Add it only if a model page lists a parameter the model ignores.

Docker users mount librechat.yaml into the container. In docker-compose.override.yml, keep the volume mount uncommented:

Configuration changes take effect after a restart: docker compose down && docker compose up -d. The endpoint then appears in the selector under the name you gave it, with the models in models.default.

Prices per million tokens and context windows, read from the live catalog when this page was built. The last column is one chat turn of 2,000 tokens in and 500 out.

Use the id exactly as the Models page prints it, not the display name. To add a model later, add its id to models.default and restart. The Claude ids answer the OpenAI format here, so one endpoint covers them with the rest.

One new chat on Claude Sonnet 5.5, first reply only: 2,000 tokens in and 500 out come to $0.0039. LibreChat sends the whole conversation on every turn, so input grows with the thread.

With titleConvo on, each new chat adds one title request of about 1,000 tokens in and 20 out: $0.0013 on Claude Sonnet 5.5, $0.000063 on Claude Haiku 5.5. Over 50 new chats a day that is $0.063 against $0.0032, which is why the example sets titleModel.

Each request shows on the console's Usage page with its tokens and charge. See how pricing works for cached input and the rest of the billing rules.

From the machine that runs LibreChat, list the models the key can call:

Then send one streamed request the way LibreChat does. -i prints the headers, including the request id.

If both work and LibreChat does not, the problem is in the YAML or the mount. The API docs list every endpoint and error.

• The endpoint is missing from the selector: the file is not mounted, or the YAML did not parse. Check the container log after docker compose up.
• 401: the key is wrong, revoked or the variable name in .env does not match the one in apiKey.
• 402: the balance is too low. Top up in the console and the next message goes through.
• 404: a model id in models.default is not in the catalog, or a display name was typed where the id belongs.
• 429: the key hit its rate limit. Wait and retry; every user of one endpoint shares one key and one limit.
• Every model in the catalog shows up: fetch is on. Set it to false and keep the list in models.default.

Field names, the baseURL convention, the env-var reference and the mount are from LibreChat's own docs, read 2026-10-31: Custom Endpoint Object Structure and Custom Endpoints quick start. Prices are from the live catalog at build time.