contextifyai 0.1.0

The simplest way to build AI agents - Built on Vercel AI SDK with PromptFlow tracking

ContextifyAI

Drop-in replacement for Vercel AI SDK with automatic tracking + editable prompts.

Change one line → Get tracking. Change one more → Get editable prompts.

🌐 Live Demo: Dashboard | API

---

The Problem

You're using Vercel AI SDK and want to:

• Track your LLM calls (latency, costs, responses)
• Edit prompts without code changes
• Let non-technical team members improve prompts

The Solution

ContextifyAI is a drop-in replacement for ai-sdk-python with 3 stages:

Stage 1: Vercel AI SDK (Your Current Code)

from ai_sdk import generate_text, anthropic

model = anthropic("claude-3-5-sonnet-20241022", api_key=api_key)
result = generate_text(model=model, prompt="What is AI?")

❌ No tracking | ❌ Hardcoded prompts

Stage 2: ContextifyAI Drop-in (1 Line Change)

from contextifyai import generate_text, anthropic  # <-- ONLY CHANGE

model = anthropic("claude-3-5-sonnet-20241022", api_key=api_key)
result = generate_text(model=model, prompt="What is AI?")

✅ Automatic tracking | ❌ Still hardcoded prompts

Stage 3: Editable Prompts (2 Line Changes Total)

from contextifyai import generate_text, anthropic  # <-- Change 1

model = anthropic("claude-3-5-sonnet-20241022", api_key=api_key)
result = generate_text(model=model, prompt_template="ai_explanation")  # <-- Change 2

✅ Automatic tracking | ✅ Edit prompts from dashboard without code changes!

---

Installation

pip install git+https://github.com/sanjevvishnu/contextifyAI.git

Or install from the package directory:

git clone https://github.com/sanjevvishnu/contextifyAI.git
cd contextifyAI
pip install -e contextifyai/

---

What You Get

Feature	Vercel AI SDK	ContextifyAI
Works	✅	✅
Same API	✅	✅
Code changes	-	1 line (import)
Tracking	❌	✅ Automatic
Dashboard	❌	✅ localhost:3001
Latency	❌	✅ Measured
Cost	❌	✅ Calculated
Prompt versioning	❌	✅ Built-in

---

Migration Guide

Step 1: Install

pip install contextifyai

Step 2: Change Import

- from ai_sdk import generate_text, anthropic
+ from contextifyai import generate_text, anthropic

Step 3: Done!

That's it. Everything else stays exactly the same.

---

Example

from contextifyai import generate_text, anthropic
import os

# Setup (same as Vercel AI SDK)
api_key = os.getenv("ANTHROPIC_API_KEY")
model = anthropic("claude-3-5-sonnet-20241022", api_key=api_key)

# All your existing code works:

# Simple
result = generate_text(model=model, prompt="Hello")

# With temperature
result = generate_text(
    model=model,
    prompt="Write a poem",
    temperature=0.9
)

# With system instruction
result = generate_text(
    model=model,
    system="You are a helpful teacher",
    prompt="Explain AI"
)

# Every call is automatically tracked!
# View at: http://localhost:3001

---

Run the Demos

🚀 Complete Production Demo (RECOMMENDED)

# Comprehensive demo explaining the entire flow
python examples/complete_demo.py

This demo:

• Makes 5 different LLM calls (customer support, code review, creative writing, data analysis, translation)
• Explains what happens in the backend server for each call
• Shows how the dashboard UI gets updated
• Verifies all calls were tracked successfully
• Includes detailed code comments explaining the entire architecture

Demo 1: Before vs After (Tracking)

# Before - no tracking
python examples/before_vercel_only.py

# After - automatic tracking (1 line changed)
python examples/after_contextifyai.py

# Open dashboard to see tracked calls
open https://promptflow-ez9ndjop6-sanjevvishnus-projects.vercel.app

Demo 2: Editable Prompts

# Run with templates
python examples/with_templates.py

# Edit template in dashboard
# Click "Prompt Templates" tab → Edit template

# Run again - see updated prompt (NO CODE CHANGES!)
python examples/with_templates.py

---

Features

🎯 Drop-in Replacement (Stage 2)

• Same API as ai-sdk-python
• 1 line change (import)
• Automatic tracking
• Dashboard at http://localhost:3001

📝 Editable Prompts (Stage 3)

• Edit prompts from dashboard
• No code deploys needed
• Non-technical users can improve prompts
• A/B testing without code changes
• Template versioning

📊 Analytics Dashboard

• View all LLM calls
• Track latency, tokens, costs
• Search and filter calls
• Edit prompt templates

---

Configuration

Environment Variables (Recommended)

# Required
ANTHROPIC_API_KEY=sk-ant-...
# or
OPENAI_API_KEY=sk-...

# Optional (use production backend)
PROMPTFLOW_URL=https://backend-black-six-59.vercel.app
PROMPTFLOW_API_KEY=pk_test_123456

# Or run locally
# PROMPTFLOW_URL=http://localhost:8000
# PROMPTFLOW_API_KEY=pk_test_123456

Programmatic (Optional)

from contextifyai import configure

# Use production backend
configure(
    promptflow_url="https://backend-black-six-59.vercel.app",
    promptflow_api_key="pk_test_123456"
)

# Or local backend
# configure(
#     promptflow_url="http://localhost:8000",
#     promptflow_api_key="pk_test_123456"
# )

---

Architecture

Your Code
    ↓
ContextifyAI (thin wrapper)
    ↓
Vercel AI SDK (ai-sdk-python)
    ↓
LLM Provider (OpenAI, Anthropic)

Tracking (parallel) →  PromptFlow Backend → Dashboard

---

FAQ

Q: Will this break my existing code?

A: No. ContextifyAI wraps Vercel AI SDK, so if your code works now, it will keep working.

Q: What if PromptFlow backend is down?

A: Tracking fails silently. Your code continues to work normally.

Q: Does this slow down my app?

A: ~5ms overhead per call for tracking. Negligible.

Q: Can I turn off tracking?

A: Yes, just switch back to from ai_sdk import ...

Q: Do I need to change my code besides the import?

A: No. Everything else stays exactly the same.

Q: What Vercel AI SDK features are supported?

A: Currently generate_text(), anthropic(), openai(). Streaming and tool calling coming soon.

---

Roadmap

• ✅ v0.1.0: Drop-in replacement for generate_text()
• 🔄 v0.2.0: Streaming support
• 🔄 v0.3.0: Tool calling support
• 🔄 v0.4.0: More providers (Cohere, Google, etc.)

---

Documentation

• DEMO_COMPARISON.md - Before/After comparison demo
• MIGRATION_GUIDE.md - Complete migration from Vercel AI SDK
• TEMPLATE_DEMO.md - Editable prompts demonstration
• CODE_COMPARISON.md - Exact code differences side-by-side

---

Repository Structure

contextifyAI/
├── contextifyai/          # Python package (pip install)
│   ├── config.py         # Configuration management
│   └── vercel_compatible.py  # Wrapped Vercel AI SDK functions
├── dashboard/            # Next.js frontend (Vercel)
│   └── app/             # Dashboard with dark mode
├── backend/              # FastAPI backend (Vercel)
│   └── app/             # Tracking API
└── examples/            # Usage examples
    ├── before_vercel_only.py
    ├── after_contextifyai.py
    └── with_templates.py

Running Locally

Backend

cd backend
pip install -r requirements.txt
python server.py
# Runs on http://localhost:8000

Dashboard

cd dashboard
npm install
npm run dev
# Runs on http://localhost:3000

License

MIT

---

Summary

✅ One line change → Full tracking

- from ai_sdk import generate_text, anthropic
+ from contextifyai import generate_text, anthropic

✅ Same API → No code changes

✅ Same results → Works exactly the same

✅ Plus tracking → Dashboard, analytics, versioning

---

Get started now:

pip install git+https://github.com/sanjevvishnu/contextifyAI.git

Then change your import and you're done! 🎉

Try the live demo: Dashboard