Coolhand Python

Monitor and log LLM API calls from OpenAI, Anthropic, Google Gemini, GitHub Copilot, Vertex AI, OpenRouter, OpenCode, Cloudflare AI Gateway, ElevenLabs, DeepSeek, Mistral, Perplexity, xAI, Cohere, Amazon Bedrock, Ollama, and more — including pydantic-ai and Dramatiq — to the Coolhand analytics platform.

Python 3.8 and 3.9 support deprecated: As of v0.4.0, coolhand requires Python 3.10 or later. Python 3.8 reached end-of-life in October 2024 and 3.9 in October 2025. If you are on an older Python version, pin to coolhand<0.4.0.

Installation

pip install coolhand

Getting Started

1. Get API Key: Visit coolhandlabs.com to create a free account
2. Install: pip install coolhand
3. Configure: Set COOLHAND_API_KEY and import coolhand in your app
4. Deploy: Your AI calls are now automatically monitored!

Quick Start

Automatic Global Monitoring

Set it and forget it! Monitor ALL AI API calls across your entire application with minimal configuration.

import coolhand  # Auto-initializes and starts monitoring

# That's it! ALL AI API calls are now automatically monitored:
# ✅ OpenAI SDK calls
# ✅ Anthropic API calls
# ✅ Google Gemini API calls
# ✅ GitHub Copilot SDK calls
# ✅ ANY library making AI API calls via httpx

# Your existing code works unchanged:
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)
# The request and response have been automatically logged to Coolhand!

Why Automatic Monitoring:

• Zero refactoring - No code changes to existing services
• Complete coverage - Monitors all AI libraries using httpx automatically
• Security built-in - Automatic credential sanitization
• Performance optimized - Negligible overhead via async logging
• Future-proof - Automatically captures new AI calls added by your team

Configuration

Environment Variables

Manual Configuration

from coolhand import Coolhand

coolhand_client = Coolhand(
    api_key='your-api-key',
    silent=False,  # Enable verbose logging
)

For excluding noisy endpoints (health checks, batch jobs) or pointing the SDK at a self-hosted backend, see Advanced Configuration.

Usage Examples

With OpenAI

import coolhand
from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)
# Request automatically logged to Coolhand!

With Anthropic

import coolhand
from anthropic import Anthropic

client = Anthropic()
response = client.messages.create(
    model="claude-3-opus-20240229",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.content[0].text)
# Request automatically logged to Coolhand!

With Google Gemini

import coolhand
import google.generativeai as genai

genai.configure(api_key="your-gemini-api-key")
model = genai.GenerativeModel("gemini-pro")
response = model.generate_content("Hello!")
print(response.text)
# Request automatically logged to Coolhand!

With GitHub Copilot SDK

import coolhand
# GitHub Copilot SDK calls are automatically intercepted
# via JsonRpcClient patching — no additional setup needed.

With Streaming

import coolhand
from openai import OpenAI

client = OpenAI()
stream = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Tell me a story"}],
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")
# Complete streamed response automatically logged to Coolhand!

What Gets Logged

The monitor captures:

• Request Data: Method, URL, headers, request body
• Response Data: Status code, headers, response body
• Timing: Request timestamp, response timestamp, duration
• LLM-Specific: Model used, token counts, streaming status

Headers containing API keys, sensitive URL query parameters, and datastore credentials embedded in request bodies (such as Azure OpenAI On Your Data) are automatically redacted. See Supported Libraries for the exact scope.

Supported Libraries

Coolhand works with any library that uses httpx, requests, or the GitHub Copilot SDK:

• OpenAI Python SDK
• Anthropic Python SDK
• Google Gemini (google-generativeai / google-genai)
• Azure OpenAI (including Azure Government and Azure China endpoints)
• Azure AI Foundry / Azure AI Inference (azure-ai-inference)
• Azure Machine Learning managed online endpoints
• GitHub Models
• GitHub Copilot SDK
• Vertex AI
• Cloudflare AI Gateway
• OpenRouter
• OpenCode
• ElevenLabs
• DeepSeek
• Mistral
• Perplexity
• xAI
• Cohere (v2 chat, v1/v2 embed)
• Amazon Bedrock
• TypeSafe Jev (System One)
• Ollama (non-localhost hosts on the default port, e.g. http://ollama:11434)
• pydantic-ai
• Any other library using httpx or requests

See Supported Libraries for the full interception mechanism breakdown and streaming details.

How It Works

Importing coolhand patches httpx, requests, and the GitHub Copilot JsonRpcClient at the class level. Requests to LLM APIs are intercepted, credentials are redacted, and data is submitted to Coolhand. Non-LLM requests pass through unchanged.

Integrations

Feedback Service

Collect user feedback on LLM responses to improve your AI outputs.

Frontend Feedback Widget: For browser-based feedback collection, see coolhand-js.

from coolhand import Coolhand

ch = Coolhand(api_key='your-api-key')

# Submit positive feedback
ch.create_feedback({
    'llm_request_log_id': 'abc123def456',  # hashid from a prior response; a raw integer FK also still works
    'sentiment': 'like',
    'explanation': 'Very helpful response!'
})

# Submit negative feedback with a correction
ch.create_feedback({
    'original_output': 'The capital of France is London.',
    'sentiment': 'dislike',
    'revised_output': 'The capital of France is Paris.'
})

Running inside an event loop (e.g. an async web handler)? Use await ch.acreate_feedback(...) instead — same fields, same result, non-blocking.

For the full field reference, matching strategies, sentiment values, and async usage, see Feedback API.

Reading Templates

search_templates and get_template read back the LLM request templates your logs are matched against. Unlike the monitoring and feedback calls, these require your private API key (the public key is write-only and 401s):

from coolhand import Coolhand

ch = Coolhand(api_key='your-private-api-key')

result = ch.search_templates(search='summar', status='published')

template = ch.get_template(result['templates'][0]['id'])

Search is a parameter on the list endpoint, not a route of its own. The Unmatched / Ignored API Calls system buckets are hidden unless you pass include_system=True, and prompt patterns come from get_template only.

These are read methods, so they raise CoolhandAPIError (carrying the HTTP status on .status) rather than returning None the way create_feedback does — a caller needs to tell a 404 from a retryable 504.

For the full filter reference, pagination, and error handling, see Reading Templates.

Linking Feedback to an Optimization

link_feedback, bulk_link_feedback (batches of 100) and unlink_feedback attach feedback to an optimization as evidence. Private API key required; errors raise CoolhandAPIError. See Linking Feedback.

Troubleshooting

Enable Debug Output

from coolhand import Coolhand

Coolhand(
    api_key='your-api-key',
    silent=False  # Enable verbose logging
)

Or via environment variable:

export COOLHAND_SILENT=false

API Key

Sign up for free at coolhandlabs.com to get your API key and start monitoring your LLM usage.

What you get:

• Complete LLM request and response logging
• Usage analytics and insights
• No credit card required to start

Security

• API keys in request headers are automatically redacted
• Sensitive URL query parameters (key, api_key, token, etc.) are automatically redacted
• Datastore credentials embedded in request bodies (e.g. Azure OpenAI On Your Data) are automatically redacted
• No sensitive data is exposed in logs
• All data is sent via HTTPS to Coolhand servers

Documentation

• Advanced Configuration — exclude patterns, self-hosted deployments, custom intercept addresses
• Feedback API — full field reference, matching strategies, sentiment values, async usage
• Supported Libraries — interception mechanisms, streaming, thread/process safety, delivery model
• Dramatiq + pydantic-ai — task queue integration guide, known gaps, workarounds
• Linking Feedback to an Optimization — link one or many feedbacks to an optimization as evidence, using the private API key
• Reading Templates — search LLM request templates and fetch a single one, prompt patterns included, using the private API key

Related Packages

• Frontend (Feedback Collection Widget): coolhand-js - Frontend feedback widget for collecting user feedback on AI outputs
• Ruby: coolhand gem - Coolhand monitoring for Ruby applications
• Node.js: coolhand-node package - Coolhand monitoring for Node.js applications

Community

• Questions? Create an issue
• Contribute? Submit a pull request
• Support? Visit coolhandlabs.com

About Coolhand Labs

Coolhand Labs builds observability and feedback tooling for AI-powered applications. Our platform helps teams monitor LLM usage, collect structured human feedback, and improve output quality over time — across every provider and framework.

License

Apache-2.0