Torrix Python SDK

Record every OpenAI and Anthropic call your Python app or agent makes in Torrix, the self-hosted observability platform for AI agents. You get the prompt, response, tokens, cost, latency and tool calls of each call, grouped into traces and sessions.

• No dependencies of its own. Works with sync, async and streaming clients.
• Sends data in a background thread, so your app is not slowed down.
• Your data goes only to your own Torrix server.

Install

pip install torrix-sdk

You also need a running Torrix server. The quickest way:

docker run -d --name torrix -p 8088:8088 -v torrix_data:/data gettorrixai/torrix:latest

Open http://localhost:8088, create your account, and copy an API key from Integrations.

Use

The package installs as torrix-sdk and imports as torrix.

import torrix

torrix.init(api_key="trxk_YOUR_KEY", base_url="http://localhost:8088")

That is all. Every call made through the installed openai and anthropic packages is recorded from now on.

Group calls into one turn and name the agent that made them:

with torrix.run_context(session_id="chat-7", agent_name="support-bot"):
    answer = client.chat.completions.create(model="gpt-4o-mini", messages=[...])

Short scripts and serverless functions should call torrix.flush() before they exit.

Learn more

• Python SDK guide
• How Torrix organizes your data
• Torrix on GitHub

MIT licensed.