diff --git a/README.md b/README.md
index ca6a1c1..ed0fdcc 100644
--- a/README.md
+++ b/README.md
@@ -41,9 +41,23 @@ service.
 - `HTTP_PORT`, default `8080`
 - `DATABASE_URL`, required
 - `MIGRATE_ON_START`, default `true`
+- `LLM_BASE_URL`, primary OpenAI-compatible LLM endpoint
+- `LLM_API_KEY`, primary LLM API key
+- `LLM_MODEL`, default `qwen2.5-14b`
+- `LLM_TIMEOUT`, default `5m`
+- `WHISPERX_URL`, WhisperX endpoint for transcription jobs
+- `OPENCLAW_URL`, optional OpenClaw gateway URL if we route through OpenClaw
+  instead of direct vLLM
 
 ## Next integration step
 
 `telephony` should first mirror low-risk analysis jobs into this service while
 continuing local processing. Remote execution can then be enabled by feature
 flag per task type.
+
+## OpenClaw note
+
+Current Portal services call the local AI server directly: vLLM for LLM tasks
+and WhisperX for transcription. OpenClaw is not required for the current
+`ai-service` queue deployment. It becomes useful if we want centralized model
+routing, provider fallback, request policy and cross-model gateway behavior.
diff --git a/internal/config/config.go b/internal/config/config.go
index c24140b..37cb844 100644
--- a/internal/config/config.go
+++ b/internal/config/config.go
@@ -3,6 +3,7 @@ package config
 import (
 	"os"
 	"strconv"
+	"time"
 )
 
 type Config struct {
@@ -10,6 +11,13 @@ type Config struct {
 	HTTPPort       int
 	DatabaseURL    string
 	MigrateOnStart bool
+
+	LLMBaseURL  string
+	LLMAPIKey   string
+	LLMModel    string
+	LLMTimeout  time.Duration
+	WhisperXURL string
+	OpenClawURL string
 }
 
 func Load() Config {
@@ -18,6 +26,13 @@ func Load() Config {
 		HTTPPort:       envInt("HTTP_PORT", 8080),
 		DatabaseURL:    envString("DATABASE_URL", ""),
 		MigrateOnStart: envBool("MIGRATE_ON_START", true),
+
+		LLMBaseURL:  envString("LLM_BASE_URL", ""),
+		LLMAPIKey:   envString("LLM_API_KEY", ""),
+		LLMModel:    envString("LLM_MODEL", "qwen2.5-14b"),
+		LLMTimeout:  envDuration("LLM_TIMEOUT", 5*time.Minute),
+		WhisperXURL: envString("WHISPERX_URL", ""),
+		OpenClawURL: envString("OPENCLAW_URL", ""),
 	}
 }
 
@@ -51,3 +66,15 @@ func envBool(key string, fallback bool) bool {
 	}
 	return v
 }
+
+func envDuration(key string, fallback time.Duration) time.Duration {
+	raw := os.Getenv(key)
+	if raw == "" {
+		return fallback
+	}
+	v, err := time.ParseDuration(raw)
+	if err != nil {
+		return fallback
+	}
+	return v
+}
diff --git a/k8s/configmap.yaml b/k8s/configmap.yaml
index 3f24174..c4b9285 100644
--- a/k8s/configmap.yaml
+++ b/k8s/configmap.yaml
@@ -7,3 +7,10 @@ data:
   HTTP_HOST: "0.0.0.0"
   HTTP_PORT: "8080"
   MIGRATE_ON_START: "true"
+  # Default direct AI endpoints. OpenClaw can replace LLM_BASE_URL later when
+  # we decide to route model traffic through a gateway instead of direct vLLM.
+  LLM_BASE_URL: "http://10.2.3.5:8002"
+  LLM_MODEL: "qwen2.5-14b"
+  LLM_TIMEOUT: "5m"
+  WHISPERX_URL: "http://10.2.3.5:8001"
+  OPENCLAW_URL: ""
diff --git a/k8s/secrets.yaml b/k8s/secrets.yaml
index 9de00a2..41b1a07 100644
--- a/k8s/secrets.yaml
+++ b/k8s/secrets.yaml
@@ -17,3 +17,4 @@ metadata:
 type: Opaque
 stringData:
   DATABASE_URL: "postgres://ai_service:ai_service@postgres:5432/ai_service?sslmode=disable"
+  LLM_API_KEY: "sk-111f838ccec43406e078cd9094b6797307cb895236179f32"