Skip to content

Worker Environment Reference ​

When you deploy a customer agent (e.g., via zelkor deploy), the platform automatically injects environment variables into the pod to wire it to the platform gateways.

Important: You must never mount real provider API keys (e.g., OpenAI or Anthropic tokens) into the agent worker environment. Provider keys belong in the platform's AI Gateway configuration.

Injected Variables ​

The following variables are provided to your agent container:

VariableExample ValueDescription
OPENAI_BASE_URLhttp://zelkor-platform-ai-gateway:80/v1Forces all OpenAI-compatible SDK traffic through the intercept plane.
OPENAI_API_KEYplatform-consumer-abc123A consumer key used by the AI Gateway for tenant rate-limiting. Not the upstream provider key.
MCP_URLhttp://zelkor-platform-mcpThe unified MCP gateway endpoint. The agent connects here to discover and call tools.
OTEL_EXPORTER_OTLP_ENDPOINThttp://zelkor-platform-langfuse:3000Points OpenTelemetry traces to the platform Langfuse instance.
LANGFUSE_HOST_HEADERlangfuse.example.comUsed for OTLP routing if required by the ingress topology.
REDIS_CHANNEL_PREFIXaegra:my-agent:run:Isolates Server-Sent Events (SSE) and queues on the shared Valkey broker.

If you are writing raw code (not using Aegra/LangGraph), ensure your LLM clients respect OPENAI_BASE_URL and OPENAI_API_KEY to inherit platform guardrails and traces.