I Built India's First AI Agent Observability Platform — Here's How to Use It (Free, Open Source)

Harsh Garg — Thu, 18 Jun 2026 11:30:00 +0000

Step 2: Add 2 Lines to Your Agent
from drishti import Drishti

drishti = Drishti(api_key="Your_API_KEY")

with drishti.trace("my_agent") as trace:
result = agent.run(user_query)
trace.set_output(result)

That's it. Every LLM call is now tracked.

What You See on the Dashboard
Live Demo: https://youtu.be/k8U_DwPs1Uk

Once integrated, your dashboard shows:

Today's Spend in ₹
Total Traces with latency
Error Rate with Hinglish explanations
Step-by-step timeline of every agent execution

Hinglish Error Example
Error: Failed to connect to shipping-api.example.com — timeout after 5s

Hinglish Explanation:
🔮 Timeout!
Request ka time limit khatam ho gaya — server ne response nahi diya.

Fix karo: Timeout value badhaao (default 30s se 60s karo).
Ya retry logic add karo apne code mein.

Pricing Built for India

Free: 10,000 traces/month, 7-day history
Pro: ₹799/month, 500K traces, WhatsApp alerts
Team: ₹2,499/month, unlimited traces

No dollar conversion. No hidden fees. INR always.

Open Source Core

The SDK is MIT licensed on GitHub. Self-host for free or use our managed cloud.
🔗 Website: drishtiai.dev
📦 GitHub: github.com/Harsh-0602/drishti-sdk

Built for Indian Developers

I'm Harsh, a solo developer from India. I built Drishti because I faced this problem every day. If you're building with LLMs in India, I'd genuinely love your feedback.

Try it free — no credit card required.

What features would you like to see next? Let me know in the comments 👇

Stop writing CLAUDE.md manually — I built a CLI that generates it from your codebase

Harsh Garg — Thu, 28 May 2026 12:30:00 +0000

If you use Claude Code, Cursor, or GitHub Copilot, you've probably written a CLAUDE.md or .cursorrules file at some point.

You carefully document your stack, conventions, and commands. Then three weeks later your project has changed, you've added Zustand, removed Redux, and your AI tool is still following the old rules.

I kept running into this. So I built CommitScope.

What it does

CommitScope scans your project and auto-generates:

CLAUDE.md for Claude Code
.cursor/rules/project.mdc for Cursor
AGENTS.md for OpenAI Codex
.github/copilot-instructions.md for GitHub Copilot

All four. One command.

How it works

npm install -g commitscope
commitscope init     # set your AI provider + API key (stored encrypted locally)
commitscope scan     # analyzes your project
commitscope generate # generates all 4 files

The scan detects your framework, language, package manager, naming conventions, dev commands — and the AI generates context files that are actually specific to your project, not generic boilerplate.

Privacy first

Your code never leaves your machine. CommitScope uses your own API key and calls your chosen provider directly. Zero telemetry. No analytics. Nothing phoned home.

This matters if you work at a company where sending code to third-party servers is a concern.

Why I built it

The alternative is copy-pasting your whole project structure into ChatGPT every time something changes. CommitScope makes it a 10-second habit instead.

It's free and open source.

npm: npm install -g commitscope
GitHub: github.com/Harsh-0602/commitscope

Would love feedback — especially if you run into issues with a specific framework or stack.

claudecode #cursor #devtools #productivity #opensource