From cb796981eeeb5c199aacdb0e11d458a4226ccebd Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 22 Jul 2026 00:20:28 -0700
Subject: [PATCH] feat: add HK LiteLLM Vertex relay

---
 servers/bwh-hk-dc3/litellm/README.md    |  5 ++++
 servers/bwh-hk-dc3/litellm/compose.yaml | 40 +++++++++++++++++++++++++
 servers/bwh-hk-dc3/litellm/config.yaml  | 15 ++++++++++
 3 files changed, 60 insertions(+)
 create mode 100644 servers/bwh-hk-dc3/litellm/README.md
 create mode 100644 servers/bwh-hk-dc3/litellm/compose.yaml
 create mode 100644 servers/bwh-hk-dc3/litellm/config.yaml

diff --git a/servers/bwh-hk-dc3/litellm/README.md b/servers/bwh-hk-dc3/litellm/README.md
new file mode 100644
index 0000000..298ec8b
--- /dev/null
+++ b/servers/bwh-hk-dc3/litellm/README.md
@@ -0,0 +1,5 @@
+# LiteLLM Vertex AI Relay
+
+- 仅通过 HK 节点的 Tailscale IPv4 地址监听；`LISTEN_IP` 不要填写公网地址或 `0.0.0.0`。
+- ADC、LiteLLM key 仅存放于 Arcane `.env`，不得提交 Git。
+- 后续 Vertex Anthropic 模型在 `config.yaml` 增加独立模型项。
diff --git a/servers/bwh-hk-dc3/litellm/compose.yaml b/servers/bwh-hk-dc3/litellm/compose.yaml
new file mode 100644
index 0000000..3d106bd
--- /dev/null
+++ b/servers/bwh-hk-dc3/litellm/compose.yaml
@@ -0,0 +1,40 @@
+services:
+  adc-init:
+    image: alpine:3.22
+    restart: "no"
+    environment:
+      GOOGLE_ADC_B64: ${GOOGLE_ADC_B64}
+    volumes:
+      - google-adc:/credentials
+    command:
+      - /bin/sh
+      - -c
+      - |
+        umask 077
+        printf '%s' "$${GOOGLE_ADC_B64}" | base64 -d > /credentials/application_default_credentials.json
+
+  litellm:
+    image: ghcr.io/berriai/litellm:main-stable
+    restart: unless-stopped
+    depends_on:
+      adc-init:
+        condition: service_completed_successfully
+    ports:
+      - "${LISTEN_IP}:${HOST_PORT:-4000}:4000"
+    environment:
+      GOOGLE_APPLICATION_CREDENTIALS: /credentials/application_default_credentials.json
+      VERTEXAI_PROJECT: ${VERTEXAI_PROJECT}
+      VERTEXAI_LOCATION: ${VERTEXAI_LOCATION:-global}
+      LITELLM_MASTER_KEY: ${LITELLM_MASTER_KEY}
+    volumes:
+      - ./config.yaml:/app/config.yaml:ro
+      - google-adc:/credentials:ro
+    command: ["--config", "/app/config.yaml", "--port", "4000"]
+    logging:
+      driver: json-file
+      options:
+        max-size: 1m
+        max-file: "2"
+
+volumes:
+  google-adc:
diff --git a/servers/bwh-hk-dc3/litellm/config.yaml b/servers/bwh-hk-dc3/litellm/config.yaml
new file mode 100644
index 0000000..b160078
--- /dev/null
+++ b/servers/bwh-hk-dc3/litellm/config.yaml
@@ -0,0 +1,15 @@
+model_list:
+  - model_name: gemini-3.6-flash
+    litellm_params:
+      model: vertex_ai/gemini-3.6-flash
+      vertex_project: os.environ/VERTEXAI_PROJECT
+      vertex_location: os.environ/VERTEXAI_LOCATION
+
+  - model_name: gemini-3.5-flash-lite
+    litellm_params:
+      model: vertex_ai/gemini-3.5-flash-lite
+      vertex_project: os.environ/VERTEXAI_PROJECT
+      vertex_location: os.environ/VERTEXAI_LOCATION
+
+general_settings:
+  master_key: os.environ/LITELLM_MASTER_KEY