DEV Community

Cover image for I Built a Local Alternative to LangSmith — 600+ Downloads in One Day 🚀 | Shivnath Tathe
Shivnath Tathe
Shivnath Tathe

Posted on

I Built a Local Alternative to LangSmith — 600+ Downloads in One Day 🚀 | Shivnath Tathe

#ai #python #agents #llm

I didn't expect this.

I published opensmith yesterday — a local-first LLM pipeline tracer
built as an alternative to LangSmith.

No cloud. No API key. No account. No data leaving your machine.

Just pip install opensmith and you're tracing.

And in one day — 600+ downloads.

Why I Built This

Every time I built an LLM app, I had to:

  • Create a LangSmith account
  • Set environment variables
  • Send my traces to someone else's cloud
  • Pray my data was safe

For production enterprise apps, healthcare tools, or anything
privacy-sensitive — that's a dealbreaker.

So I asked: what if observability just worked locally?

Why Observability Matters

You cannot improve what you cannot see.

When your LLM pipeline breaks, you need to know:

  • 🔍 Which prompt caused the bad output?
  • ⏱️ Where is the latency spike?
  • 💸 Which call is burning tokens unnecessarily?
  • 🔁 Which chain step is failing silently?

Without tracing, you're flying blind.

LangSmith solves this — but at the cost of your data going
to their servers.

opensmith solves it without that cost.

What opensmith Does 

from opensmith import trace

@trace
def my_llm_pipeline(prompt):
    # your LLM calls here
    return response
Enter fullscreen mode Exit fullscreen mode

That's it. One decorator.

  • ✅ Traces every LLM call locally
  • ✅ No cloud dependency
  • ✅ No setup beyond pip install
  • ✅ Works with any LLM — OpenAI, Anthropic, local models
  • ✅ MIT licensed, fully open source

Local-First Is Not a Compromise

People assume local = limited.

I'd argue the opposite.

Local-first means:

  • Your traces never leave your machine
  • No rate limits on observability
  • Works fully offline
  • Zero vendor lock-in
  • Instant setup in any environment

This is especially critical for:

  • 🏥 Healthcare AI apps
  • 🏦 Banking and fintech LLM tools
  • 🛡️ Defense and enterprise systems
  • 👨‍💻 Solo developers who just want to ship fast

The Response

600+ downloads. One day. Zero marketing.

That number tells me one thing — developers are tired of
giving their data to yet another SaaS just to debug their
own code.

Try It 

pip install opensmith
Enter fullscreen mode Exit fullscreen mode

GitHub: github.com/shivnathtathe/opensmith

PyPI: pypi.org/project/opensmith

If you're building LLM apps and care about privacy,
observability, or just hate unnecessary signups —
give it a try.

And if you find it useful, a ⭐ on GitHub means
the world to an independent researcher building
in public. 🙏

Built by an independent AI researcher in Hyderabad, India.
Believer in AI that runs without expensive infrastructure.

Top comments (0)