Using Downsizing with LibreChat

Prerequisites

A LibreChat deployment you can edit the configuration of (Docker Compose or source install) and a Downsizing account that can make requests (pay-per-use or a connected subscription both work). Grab your Inference ID and API key from the Inference API page.

LibreChat is a self-hosted server (typically run via Docker Compose alongside MongoDB and Meilisearch), not a local desktop app — so unlike CLI tools, the dsCLI can't reach into its config from your machine. Wiring it up is a one-time edit to a YAML file that lives with your LibreChat deployment.

Manual Setup

LibreChat lets you declare custom model endpoints in a librechat.yaml file at the root of your LibreChat project (create it if it doesn't exist — LibreChat ships a librechat.example.yaml you can copy). Add a custom endpoint with provider: 'anthropic' so LibreChat speaks the native Anthropic Messages API to Downsizing:

version: 1.2.8
cache: true
endpoints:
  custom:
    - name: 'Downsizing'
      provider: 'anthropic'
      apiKey: '${DOWNSIZING_API_KEY}'
      baseURL: 'http://localhost:3001/YOUR_INFERENCE_ID/anthropic'
      models:
        default: ['claude-opus-4-8']
        fetch: false
      titleConvo: true
      titleModel: 'claude-opus-4-8'
      modelDisplayLabel: 'Downsizing'

Replace YOUR_INFERENCE_ID with your workspace UUID from the dashboard. Then add the referenced API key to the .env file LibreChat loads (same directory as docker-compose.yml):

DOWNSIZING_API_KEY=dws_your_api_key_here

If you're running via Docker Compose and librechat.yamlisn't already mounted into the container, add a bind mount in docker-compose.override.yml:

services:
  api:
    volumes:
      - type: bind
        source: ./librechat.yaml
        target: /app/librechat.yaml

Restart LibreChat (docker compose restart or docker compose up -d) — like most of its configuration, librechat.yaml is only read on startup.

Verify

Open LibreChat, select Downsizing from the endpoint/model picker, and send a test message. If you get a response, the endpoint is wired up correctly. Then check your savings dashboard to see the request appear in your usage metrics.