创建 start.sh

start.sh

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# ── 1. Validate required LiteLLM vars ─────────────────────────────────────────
+: "${LITELLM_API_KEY:?LITELLM_API_KEY is required}"
+: "${LITELLM_MODEL:?LITELLM_MODEL is required (e.g. openai/gpt-4o or anthropic/claude-3-5-sonnet-20241022)}"
+
+# ── 2. Generate LiteLLM proxy config ──────────────────────────────────────────
+LITELLM_CONFIG=/tmp/litellm_config.yaml
+
+cat > "$LITELLM_CONFIG" << YAML
+model_list:
+  - model_name: default
+    litellm_params:
+      model: "${LITELLM_MODEL}"
+      api_key: "${LITELLM_API_KEY}"
+$(if [ -n "${LITELLM_API_BASE:-}" ]; then echo "      api_base: \"${LITELLM_API_BASE}\""; fi)
+
+litellm_settings:
+  drop_params: true
+  num_retries: 3
+  request_timeout: 120
+YAML
+
+echo "[start.sh] LiteLLM config written to $LITELLM_CONFIG"
+
+# ── 3. Start LiteLLM proxy in the background ──────────────────────────────────
+litellm --config "$LITELLM_CONFIG" --port 4000 --host 127.0.0.1 &
+LITELLM_PID=$!
+echo "[start.sh] LiteLLM proxy started (pid $LITELLM_PID)"
+
+# ── 4. Wait for LiteLLM to be healthy ─────────────────────────────────────────
+MAX_WAIT=30
+WAITED=0
+until curl -sf http://127.0.0.1:4000/health/liveliness > /dev/null 2>&1; do
+    if [ "$WAITED" -ge "$MAX_WAIT" ]; then
+        echo "[start.sh] ERROR: LiteLLM did not become healthy within ${MAX_WAIT}s"
+        exit 1
+    fi
+    sleep 1
+    WAITED=$((WAITED + 1))
+done
+echo "[start.sh] LiteLLM is healthy after ${WAITED}s"
+
+# ── 5. Start OpenClaw pointing to the LiteLLM proxy ──────────────────────────
+# We expose LiteLLM as an OpenAI-compatible endpoint.
+# OPENCLAW_DEFAULT_MODEL must match the model_name we set above ("default").
+exec env \
+    OPENCLAW_DEFAULT_MODEL=default \
+    OPENAI_API_KEY=litellm-proxy \
+    OPENAI_BASE_URL=http://127.0.0.1:4000 \
+    node openclaw.mjs gateway --allow-unconfigured