Unified Distributed Trace Tree
Native gateway spans and client OTLP spans are reconciled into a single distributed trace tree.
OTLP & Observability Integration
Standard OTLP/HTTP JSON and Protobuf trace contracts allow seamless integration with your existing Grafana, Datadog, Honeycomb, and Jaeger monitoring stacks.
Standard Ingest & Export
Full support for standard OTLP/HTTP JSON and Protobuf export by trace, prompt ID, or workspace.
APM Ecosystem Native Integration
Directly integrate with Grafana Tempo, Datadog, Honeycomb, and Jaeger without custom adapters.
Semantic Conventions
OpenTelemetry GenAI Semantic Attributes
Standardized trace span attributes automatically injected on every BatchIn inference call.
| Attribute | Type | Example | Description |
|---|---|---|---|
| gen_ai.system | string | "batchin" | Identifies the AI gateway system |
| gen_ai.request.model | string | "deepseek-ai/DeepSeek-V4-Pro" | Target model identifier requested |
| gen_ai.usage.prompt_tokens | int | 412 | Number of input tokens consumed |
| gen_ai.usage.completion_tokens | int | 185 | Number of completion tokens generated |
| gen_ai.response.latency_ms | float | 624.5 | Total gateway turnaround latency (ms) |
| gen_ai.response.ttft_ms | float | 92.0 | Time to first token generation (ms) |
| gen_ai.vaas.record_id | string | "rec_98bf12" | Cryptographic VaaS audit record ID |
| gen_ai.vaas.merkle_root | string | "0x3a4f...e109" | Base L2 Merkle tree batch commitment root |
| gen_ai.agent.run_id | string | "run_prod_agent_77" | Autonomous agent multi-step run ID |
| gen_ai.agent.step_id | string | "step_tool_fallback_03" | Step identifier within the agent loop |
Production Blueprint
OpenTelemetry Collector Configuration
Copy-pasteable configuration for forwarding BatchIn traces to Grafana Tempo and Datadog.
OTLP gRPC: 4317 | HTTP: 4318
# otel-collector-config.yaml
# Production OpenTelemetry Collector for BatchIn Inference Gateway
receivers:
otlp:
protocols:
grpc:
endpoint: 0.0.0.0:4317
http:
endpoint: 0.0.0.0:4318
processors:
batch:
timeout: 1s
send_batch_size: 256
memory_limiter:
check_interval: 1s
limit_percentage: 80
spike_limit_percentage: 20
exporters:
# 1. Export to Grafana Tempo / Honeycomb
otlp/tempo:
endpoint: tempo.internal:4317
tls:
insecure: true
# 2. Export to Datadog LLM Observability
datadog:
api:
key: "${DD_API_KEY}"
site: datadoghq.com
service:
pipelines:
traces:
receivers: [otlp]
processors: [memory_limiter, batch]
exporters: [otlp/tempo, datadog]