Connect LibreChat to CheapTokens: librechat.yaml custom endpoint

Your LibreChat is running, but the model menu only shows what it shipped with. To add CheapTokens you write one endpoints.custom entry in librechat.yaml: a name, the base URL https://api.cheaptokens.dev/v1, the key as an environment variable and a list of model ids. Restart the container, and a "CheapTokens" endpoint appears in the menu.

The field names below come from LibreChat's custom endpoint reference (docs "Latest (dev)" branch, checked 30 October 2026). The page carries no version number, so if your install is older and a field behaves differently, compare with the docs for your release. The API side is the one from our docs: OpenAI Chat Completions format, one balance for every model.

What the entry needs#

Four fields are required, and each one has a plain job:

• name — what the endpoint is called in the menu.
• apiKey — the key. The docs recommend a reference like ${VAR} instead of the key itself, so the secret stays out of the YAML.
• baseURL — where requests go: https://api.cheaptokens.dev/v1.
• models.default — the list of model ids shown in the menu.

Two optional fields are worth setting. titleConvo: true lets LibreChat name your chats (titles are created unless it is false), and modelDisplayLabel is the label next to the answers (the default is "AI"). models.fetch is false by default, so LibreChat will not ask the API for a model list; the menu comes from models.default. That is why the ids below are spelled out in full.

Connect it in three steps#

services:
  api:
    environment:
      CHEAPTOKENS_API_KEY: ct_your_key_here

endpoints:
  custom:
    - name: 'CheapTokens'
      apiKey: '${CHEAPTOKENS_API_KEY}'
      baseURL: 'https://api.cheaptokens.dev/v1'
      models:
        default:
          - 'zai-org/GLM-5.3-Flash'
          - 'deepseek-ai/DeepSeek-V4-Flash-0731'
          - 'MiniMaxAI/MiniMax-M2.7'
      titleConvo: true
      modelDisplayLabel: 'CheapTokens'

services:
  api:
    environment:
      CHEAPTOKENS_API_KEY: ct_your_key_here
    volumes:
      - ./librechat.yaml:/app/librechat.yaml

docker compose up -d

Open LibreChat in the browser, start a new chat and pick CheapTokens in the endpoint menu. The three models are in the list. Send "Say hi in one sentence" and you should get an answer back. If you don't, jump to the last section.

Three models in one menu#

This is the part that makes a self-hosted chat worth having. Say you run it for a small team. Most questions are everyday ones: rewrite this email, explain this error. For those, zai-org/GLM-5.3-Flash is enough, with a 390K context window and up to 8,192 tokens per answer. When someone pastes a long document and wants a long reply, they switch the same chat to deepseek-ai/DeepSeek-V4-Flash-0731 (380K context, up to 32,768 tokens per answer). MiniMaxAI/MiniMax-M2.7 (180K context) is the third opinion when two answers disagree. No second subscription, no second key: all three are billed from one balance.

What does that chat cost? Take a team of five sending 200 messages a day, each carrying about 3,000 input tokens (the history grows) and getting about 500 tokens back:

In the table, each model's price for those 200 requests is counted with input and output together, and in total it comes out cheaper than OpenRouter. Pick the row that matches your usage and compare it against the pricing page.

If the endpoint doesn't show up or answers with an error#

Most problems are one of three things. Pick the one that matches what you see:

Once the first message goes through, you can add more ids to models.default whenever you want another model in the menu. Prefer a different front end? The same idea, a base URL plus a key, is in Connect Open WebUI to CheapTokens, and the general picture is in OpenAI-compatible API: what base_url is. The models themselves are on the GLM page.