Add Dolphin LLM to LibreChat as a custom endpoint

The endpoint block

Open librechat.yaml in the LibreChat folder (copy librechat.example.yaml if you do not have one yet) and add this under endpoints:

version: 1.2.1
endpoints:
  custom:
    - name: "Dolphin LLM"
      apiKey: "${DOLPHIN_API_KEY}"
      baseURL: "https://dolphinllm.com/v1"
      models:
        default: ["uncensored"]
        fetch: false
      titleConvo: true
      titleModel: "uncensored"
      modelDisplayLabel: "Dolphin LLM"

Use the version value your LibreChat release expects (it is at the top of librechat.example.yaml). fetch: false keeps the list to the one model; with fetch: true LibreChat asks GET /v1/models, which returns the same single model.

Put the key in .env

# .env
DOLPHIN_API_KEY=YOUR_KEY

Create the key in your account (sign up first if you have none; $0.50 of trial credit is added when you first open the account). Never commit .env to git.

If every user of your install should pay from their own balance, set apiKey: "user_provided" instead. LibreChat then asks each user for a key the first time they pick Dolphin LLM, and stores it for them.

Docker: mount the file and restart

With the standard Docker setup, LibreChat reads librechat.yaml only if it is mounted into the API container. Add an override file next to docker-compose.yml:

# docker-compose.override.yml
services:
  api:
    volumes:
      - type: bind
        source: ./librechat.yaml
        target: /app/librechat.yaml

docker compose down
docker compose up -d

Open LibreChat, choose Dolphin LLM in the endpoint menu and start a chat. If the endpoint does not show up, the API container log says why, usually a YAML indentation error.

Options worth knowing

• titleConvo makes one short extra request per new conversation to name it. On a trial key (2 parallel requests, 60 a minute) set it to false if you see delays.
• Context: the model accepts up to 262,144 tokens, prompt plus reply. LibreChat sends the conversation history with each message, so long threads cost more per turn; a turn that fills the window costs about $0.0393 in input.
• Parameters: temperature, top_p, max tokens, stop and the penalties from LibreChat's parameter panel are passed through. The API is text only, so file uploads that need vision or embeddings will not work with this endpoint.
• Presets and agents: you can save a system prompt as a preset. Tool calling is supported by the API (tools, tool_choice), so LibreChat agents can use it; test one tool before you rely on it.

Shared install, one balance

With one key in .env, every LibreChat user spends from the same balance. Turn off open registration if the install is public, and check the balance in your account or the X-Balance-USD response header. Limits apply per key: after the first top-up, 8 parallel requests and 300 a minute shared by all users.

The model is tuned to answer without content refusals for lawful adult use, so the house rules are yours to set, for example in a preset system prompt. Sexual content involving minors is never allowed on the API.

Cost

At $0.15 per 1M input and $0.50 per 1M output tokens, a turn with 3,000 tokens of history and a 600-token answer costs $0.00075. A small team sending 500 such messages a day spends about $11 a month. See the pricing page for packs and bonuses.