librechat.yaml

# LibreChat custom endpoint configuration
#
# Routes LLM requests through the Stihia proxy.
# The proxy forwards them to the real provider API while applying
# Stihia threat-detection guardrails in parallel.
#
# Each endpoint sets:
#   X-Upstream-Base-URL  — the real provider base URL.
#   X-User-ID            — the LibreChat user who sent the message.
#   X-Conversation-ID    — the current LibreChat conversation.
#   X-Message-ID         — the current message ID.
#
# Note: The Stihia process key (model name or agent ID) is extracted
# directly from the request body by the proxy — no header needed.

version: "1.1.7"
cache: true

endpoints:
  custom:
    # ---- OpenAI via Stihia Proxy ----
    - name: "OpenAI"
      apiKey: "${OPENAI_API_KEY}"
      baseURL: "http://stihia-proxy:4005/v1"
      models:
        default: ["gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano", "gpt-5.3-codex"]
        fetch: false
      titleConvo: true
      titleModel: "gpt-5.4"
      summarize: false
      forcePrompt: false
      modelDisplayLabel: "OpenAI"
      headers:
        X-Upstream-Base-URL: "https://api.openai.com"
        X-User-ID: "{{LIBRECHAT_USER_ID}}"
        X-Conversation-ID: "{{LIBRECHAT_BODY_CONVERSATIONID}}"
        X-Message-ID: "{{LIBRECHAT_BODY_MESSAGEID}}"