continua-sdk 0.1.0

Python SDK for Continua agent observability and debugging

Continua SDK for Python

Python SDK for sending AI agent traces, spans, sessions, events, and engine control requests to Continua.

Installation

pip install continua-sdk

Quick Start

Initialize the client once at process startup:

from continua import Continua, span, trace

Continua.init(
    api_key="your-api-key",
    endpoint="https://api.your-continua-host.com",
    ingest_mode="server_default",  # or "sync", "async_v2"
)


@trace(name="answer_question")
def answer_question(question: str) -> str:
    with span("retrieve_context", kind="tool") as s:
        s.set_input({"question": question})
        context = "retrieved context"
        s.set_output({"context": context})

    with span("call_model", kind="llm") as s:
        s.set_input({"question": question, "context": context})
        answer = "model answer"
        s.set_output({"answer": answer})
        return answer

For a local or self-hosted Continua server, point the SDK at your server URL:

from continua import Continua

Continua.init(
    api_key="dev-api-key",
    endpoint="http://localhost:8080",
)

This package is the client library only. It sends data to a hosted or self-hosted Continua server; it does not install or run the Go server, Postgres, River workers, or debugger UI.

Async Ingest Modes

• server_default: defer to the server rollout setting.
• sync: send sync=true and wait for inline processing.
• async_v2: opt into true async ingest with X-Continua-Async-Version: 2.

True async is not read-after-write. If your code needs the batch to be fully processed before it reads the ingested data back, use ingest_mode="sync" or poll with wait_for_batch():

result = client.wait_for_batch(batch_id, timeout=30, poll_interval=0.5)

Development

# From sdks/python
uv sync

uv run pytest

uv run mypy src/

uv build

Publishing

The Python package is published from this directory:

cd sdks/python
uv build
uv publish

Prefer PyPI Trusted Publishing in CI for repeatable releases once the project is ready for public distribution.