Use environment variables

Author: Naragod

.env.example

@@ -0,0 +1,14 @@
+# Redis Configuration
+REDIS_URL=redis://localhost:6379
+
+# Ollama/Llama Server Configuration
+LLAMA_SERVER_URL=http://localhost:11434
+OLLAMA_HOST=http://localhost:11434  # Used by ollama Python library
+DEFAULT_MODEL=deepseek-coder-v2:latest
+
+# OpenTelemetry Configuration
+OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+
+# Optional: Llama.cpp CLI configuration (for local llama.cpp usage)
+# LLAMA_CPP_CLI=/data1/llama.cpp/bin/llama-cli
+# GGUF_DIR=/data1/GGUF

ai_server/server.py

@@ -26,15 +26,18 @@
 # Load environment variables from .env file
 load_dotenv()
 
+# OpenTelemetry endpoint configuration
+OTEL_EXPORTER_OTLP_ENDPOINT = os.getenv('OTEL_EXPORTER_OTLP_ENDPOINT', 'http://localhost:4317')
+
 # Configure OpenTelemetry - Shared Resource (identifies this service)
 resource = Resource.create({"service.name": "ai-server"})
 
 # ========== TRACES CONFIGURATION ==========
 # TracerProvider: Factory for creating tracers (for distributed tracing)
 tracer_provider = TracerProvider(resource=resource)
 
-# OTLP Trace Exporter: Sends traces to collector at localhost:4317
-otlp_trace_exporter = OTLPSpanExporter(endpoint="http://localhost:4317", insecure=True)
+# OTLP Trace Exporter: Sends traces to collector
+otlp_trace_exporter = OTLPSpanExporter(endpoint=OTEL_EXPORTER_OTLP_ENDPOINT, insecure=True)
 span_processor = BatchSpanProcessor(otlp_trace_exporter)
 tracer_provider.add_span_processor(span_processor)
 
@@ -43,8 +46,8 @@
 tracer = trace.get_tracer("ai-server.tracer")
 
 # ========== METRICS CONFIGURATION ==========
-# OTLP Metric Exporter: Sends metrics to collector at localhost:4317
-otlp_metric_exporter = OTLPMetricExporter(endpoint="http://localhost:4317", insecure=True)
+# OTLP Metric Exporter: Sends metrics to collector
+otlp_metric_exporter = OTLPMetricExporter(endpoint=OTEL_EXPORTER_OTLP_ENDPOINT, insecure=True)
 
 # PeriodicExportingMetricReader: Collects and exports metrics every 10 seconds
 metric_reader = PeriodicExportingMetricReader(
@@ -282,7 +285,7 @@ def authenticate() -> str:
 @app.route('/chat', methods=['POST'])
 def chat():
     """Handle chat request with optional llama_mode and system prompt parameters."""
-    # authenticate()
+    authenticate()
     model = request.form.get('model', DEFAULT_MODEL)
     content = request.form.get('content', '')
     llama_mode = request.form.get('llama_mode', 'cli')