Skip to main content

​
Overview

Dakora integrates with popular agent and orchestration frameworks. Send OpenTelemetry traces via OTLP/HTTP or log executions directly — no SDK lock‑in. Use project‑scoped APIs and your X-API-Key.

​
Microsoft Agent Framework (MAF)

Use the built-in MAF middleware for one-line setup — no manual HTTP calls required. 
pip install "dakora-client[maf]"

from dakora_client import Dakora
from dakora_agents.maf import DakoraIntegration
from agent_framework.openai import OpenAIChatClient  # or agent_framework.azure.AzureOpenAIChatClient

# 1) Initialize Dakora (uses DAKORA_API_KEY env var if present)
dakora = Dakora()

# 2) One-line observability setup (OpenTelemetry via OTLP/HTTP → Dakora, budget checks, etc.)
middleware = DakoraIntegration.setup(dakora)

# 3) Plug into your MAF chat client
client = OpenAIChatClient(api_key="sk-...", middleware=[middleware])
agent = client.create_agent(id="support-bot")

# 4) Render a Dakora template and convert to an agent message (auto-tracked)
tmpl = await dakora.prompts.render(
    "faq_responder",
    {"question": "How do I reset my password?", "knowledge_base": "Click Forgot Password."}
)
response = await agent.run(tmpl.to_message())

# Optional: flush OpenTelemetry exporter (OTLP) before exiting
DakoraIntegration.force_flush()
What you get automatically:
  • Token, latency, and cost tracking via OpenTelemetry (OTLP exporter)
  • Template attribution (via metadata markers or template_usages)
  • Budget checks (configurable TTL cache)
  • Rich timeline and hierarchy views in the UI

​
OpenAI Agents (Coming Soon)

Developer‑friendly adapter for one‑line instrumentation. No manual HTTP.

​
LangChain (Coming Soon)

Lightweight package for easy tracing and budgets with minimal setup.

​
Best Practices

  • Prefer OpenTelemetry (OTLP) ingestion for rich timelines and low overhead.
  • Use stable trace_id for idempotency.
  • Include provider, model, tokens_in, tokens_out, and latency_ms for cost visibility.
  • Link templates via template_usages for accurate attribution.

​
See Also

  • Observability & Logging: /features/logging
  • REST API: /api-reference/rest