Python SDK

The ModelTrack Python SDK auto-instruments Anthropic and OpenAI calls with a single import. No code changes required.

Installation

For now, copy the SDK file into your project. A pip install package is coming soon.

terminal

# Copy from the ModelTrack repo
cp modeltrack/sdks/python/modeltrack.py your-project/

# Coming soon:
# pip install modeltrack

Auto-instrumentation

Import modeltrackat the top of your app. That's it. All Anthropic and OpenAI SDK instances will automatically route through the ModelTrack proxy and include attribution headers.

app.py

import modeltrack  # Add this one line
import anthropic

# This client now routes through ModelTrack automatically
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}]
)

# OpenAI works the same way
import openai
client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}]
)

How it works: On import, the SDK monkey-patches anthropic.Anthropic.__init__ and openai.OpenAI.__init__ to set base_url to the proxy and inject X-ModelTrack-* headers. If you explicitly set base_urlin the constructor, it won't be overridden.

Manual base URL

If you prefer not to use auto-instrumentation, set the base URL manually:

manual.py

import anthropic

client = anthropic.Anthropic(base_url="http://localhost:8080")
response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}]
)

# For OpenAI:
import openai
client = openai.OpenAI(base_url="http://localhost:8080")

Configuration

Configure the SDK either via environment variables or programmatically with modeltrack.configure().

Environment variables

Variable	Default	Description
`MODELTRACK_PROXY_URL`	`http://localhost:8080`	Proxy URL
`MODELTRACK_TEAM`	-	Team name for attribution
`MODELTRACK_APP`	-	Application name
`MODELTRACK_FEATURE`	-	Feature name
`MODELTRACK_CUSTOMER_TIER`	-	Customer tier (e.g., enterprise, free)
`MODELTRACK_SESSION_ID`	-	Session ID for multi-step workflows
`MODELTRACK_TRACE_ID`	-	Trace ID for distributed tracing
`MODELTRACK_PROMPT_TEMPLATE`	-	Prompt template ID for cost analysis

Programmatic configuration

config.py

import modeltrack

modeltrack.configure(
    proxy_url="http://localhost:8080",
    team="ml-research",
    app="chatbot",
    feature="customer-support",
    customer_tier="enterprise",
    prompt_template="support-v2",
)

Session tracking

Use the session() context manager to group multiple LLM calls into a single session. This is useful for multi-step agent workflows where you want to track the total cost of a conversation or task.

sessions.py

import modeltrack
import anthropic

client = anthropic.Anthropic()

# All calls within this block share the same session ID
with modeltrack.session("user-query-123"):
    # Step 1: Classify the query
    classification = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=100,
        messages=[{"role": "user", "content": "Classify: refund request"}]
    )

    # Step 2: Generate response
    response = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
        messages=[{"role": "user", "content": "Handle refund for order #123"}]
    )

# With optional trace ID for distributed tracing
with modeltrack.session("session-123", trace_id="trace-456"):
    response = client.messages.create(...)

Sessions are thread-safe — each thread gets its own session context via thread-local storage.

Setting team, app, and feature

You can set attribution at three levels:

attribution.py

# 1. Environment variables (set once, applies globally)
#    MODELTRACK_TEAM=ml-research
#    MODELTRACK_APP=chatbot
#    MODELTRACK_FEATURE=customer-support

# 2. Programmatic (set once in your app startup)
import modeltrack
modeltrack.configure(team="ml-research", app="chatbot")

# 3. Per-request headers (maximum flexibility)
client = anthropic.Anthropic(
    default_headers={
        "X-ModelTrack-Team": "ml-research",
        "X-ModelTrack-App": "chatbot",
        "X-ModelTrack-Feature": "customer-support",
    }
)

Compatibility

The Python SDK auto-patches these libraries:

✓anthropic — Anthropic Python SDK
✓openai — OpenAI Python SDK

For frameworks like LangChain, CrewAI, and LlamaIndex, see the Frameworks guide.