Dakora Instrumentation

OTLP backend for LLM observability

Dakora accepts OpenTelemetry traces via OTLP/HTTP. This package provides:

1. Generic helper - Thin wrapper to configure OTEL exporters pointing to Dakora
2. Framework integrations - Batteries-included support for Microsoft Agent Framework (MAF)
   • LangChain, CrewAI, and more coming soon

Features

✅ OTLP Backend - Point your existing OTEL setup at Dakora
✅ Multi-Export - Send traces to Dakora + Jaeger/Grafana/Azure Monitor
✅ BYO OpenTelemetry - Use your own OTEL versions, no dependency conflicts
✅ Framework Integrations - Batteries-included MAF support
✅ Template Linkage - Track Dakora prompts used in executions
✅ Budget Enforcement - Pre-execution checks with caching (MAF integration)

---

Architecture

Your App → OTEL SDK → Instrumentation → OTLP Exporter → Dakora API
                                                    ↓
                                            /api/v1/traces
                                         (with X-API-Key header)

---

Quick Start

1. With OpenTelemetry Collector (Recommended)

Already running a collector? Just add Dakora as an exporter:

# otel-collector-config.yaml
exporters:
  otlphttp/dakora:
    endpoint: ${DAKORA_BASE_URL}/api/v1/traces
    headers:
      X-API-Key: ${DAKORA_API_KEY}

service:
  pipelines:
    traces:
      exporters: [otlphttp/dakora, jaeger, ...] # Multi-export!

2. Direct Integration (Python)

Step 1: Install OpenTelemetry + Instrumentations

# Core OTEL
pip install opentelemetry-sdk opentelemetry-exporter-otlp-proto-http

# Provider instrumentations (choose what you need)
pip install opentelemetry-instrumentation-openai
pip install opentelemetry-instrumentation-anthropic

# Dakora helper (optional convenience)
pip install dakora-instrumentation

Step 2: Instrument Your Providers

# Do this once at application startup
from opentelemetry.instrumentation.openai import OpenAIInstrumentor
from opentelemetry.instrumentation.anthropic import AnthropicInstrumentor

OpenAIInstrumentor().instrument()
AnthropicInstrumentor().instrument()

Step 3: Configure Dakora Exporter

from dakora import Dakora
from dakora_instrumentation import setup_instrumentation

dakora = Dakora(api_key="dk_proj_...")
setup_instrumentation(dakora, service_name="my-app")

# That's it! All instrumented SDKs now export to Dakora

2. With Microsoft Agent Framework (Batteries Included)

pip install 'dakora-instrumentation[maf]'

from dakora import Dakora
from dakora_instrumentation.frameworks.maf import DakoraIntegration
from agent_framework.azure import AzureOpenAIChatClient

# Initialize Dakora
dakora = Dakora(api_key="dk_proj_...")

# One-line OTEL setup
middleware = DakoraIntegration.setup(dakora)

# Use with any MAF client
azure_client = AzureOpenAIChatClient(
    endpoint=...,
    deployment_name=...,
    api_key=...,
    middleware=[middleware],
)

agent = azure_client.create_agent(
    id="chat-v1",
    name="ChatBot",
    instructions="You are helpful.",
)

response = await agent.run("Hello!")

---

Package Structure

dakora_instrumentation/
├── frameworks/maf/          # Microsoft Agent Framework integration
├── generic.py               # Generic OTEL setup helper
└── _internal/               # Internal utilities (private)

Public API:

from dakora_instrumentation import setup_instrumentation
from dakora_instrumentation.frameworks.maf import DakoraIntegration

---

Examples

See examples/ directory for progressive examples:

• 01_quickstart/ - Start here: basic setup and template usage
• 02_providers/ - BYO OTEL patterns (OpenAI, Anthropic, multi-provider)
• 03_maf_agents/ - MAF agent patterns (simple, tools, templates)
• 04_maf_multi_agent/ - Multi-agent orchestration (sequential, parallel, workflows)
• 05_advanced/ - Production patterns (dual export, budget checking, custom attributes)

---

FAQ

Q: Do I need to install dakora-instrumentation?

No! If you already have OTEL configured, point your exporter at Dakora's OTLP endpoint: ${DAKORA_BASE_URL}/api/v1/traces with X-API-Key header. This package provides convenience helpers.

Q: How do I install provider instrumentations?

Install them directly from the opentelemetry-instrumentation packages:

pip install opentelemetry-instrumentation-openai
pip install opentelemetry-instrumentation-anthropic

Q: When should I use which integration?

• Generic setup_instrumentation() - For direct SDK calls (OpenAI, Anthropic, etc.)
• MAF Integration - For Microsoft Agent Framework agents (batteries included)

Q: Do I need to instrument providers manually?

Yes, OpenTelemetry instrumentation is global:

OpenAIInstrumentor().instrument()  # Do once at startup

Q: Can I send traces to multiple backends?

Yes! Use additional_exporters parameter:

from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter

jaeger_exporter = OTLPSpanExporter(endpoint="http://localhost:4317")
setup_instrumentation(dakora, additional_span_exporters=[jaeger_exporter])

Q: Does this work with LangChain/CrewAI?

Not yet, but support is coming! Use generic setup_instrumentation() for now.

---

Links

• Documentation: https://docs.dakora.io
• GitHub: https://github.com/dakora-labs/dakora
• Issues: https://github.com/dakora-labs/dakora/issues
• Support: support@dakora.io

---

License

Apache License 2.0