Documentation
Niitaka SDK
Niitaka gives you full visibility into your AI agents — every LLM call, tool invocation, and decision — and a control plane to experiment, version, and enforce policies without changing your agent code.
Install
Requires Python 3.9+
bash
pip install "niitaka-sdk[openai]"See the Providers page for provider-specific install extras.
In 10 seconds
Configure once, instrument your provider, wrap your run.
python
import niitaka
import os
niitaka.configure(
api_key=os.getenv("NIITAKA_API_KEY"),
api_url="https://api.niitaka.ai",
)
niitaka.instrument_openai()
with niitaka.start_session(goal="Summarise report", agent_id="my-agent"):
response = openai_client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Summarise this report..."}],
)Every call inside start_session is automatically traced. No manual logging needed.
What's included
Observability
Session dashboards, cost & error tracking
Tracing
Full LLM + tool event timeline
Experiments
A/B test models, prompts, and guardrails
Evaluation
Replay traces against new policies
Versioning
Draft → Active → Archived lifecycle
Guardrails
Cost limits, retries, and fallbacks