Show HN: OpenSmith – Trace LLM pipelines locally with trace, no cloud

The open-source, local-first alternative to LangSmith.

opensmith is to LangSmith what Ollama is to OpenAI — the local-first, privacy-first alternative.

LangSmith	opensmith
Setup	Cloud account required	pip install opensmith
Data privacy	Sends traces to cloud	100% local, SQLite only
Framework	Best with LangChain	Works with any Python code
Cost	Free tier then paid	Free forever, open source
Offline	No	Yes
Docker	No	No
Dashboard	Hosted	localhost:7823

LangSmith is powerful, but it is built around cloud-hosted tracing and is most natural inside the LangChain ecosystem. opensmith is a local-first alternative: install it with pip

, use it with any Python LLM pipeline, and inspect traces on your machine without accounts, hosted services, Docker, or configuration. No trace data leaves your machine.

pip install opensmith

Optional integrations:

pip install "opensmith[otel]"
pip install "opensmith[postgres]"
pip install "opensmith[all]"

Use opensmith[all]

to install both OpenTelemetry and Postgres support.

from opensmith import trace

@trace
def call_llm(prompt: str):
    return openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
    )

@trace
def my_pipeline(question: str):
    docs = search_docs(question)
    return call_llm(docs + question)

Async functions are supported:

from opensmith import trace

@trace(tags=["production", "rag"])
async def call_llm(prompt: str):
    return await openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
    )

Warn when a trace exceeds a token budget:

from opensmith import trace

@trace(token_budget=1000)
def my_pipeline():
    return call_llm("summarize this document")
python
from opensmith import trace

with trace("my_pipeline", tags=["debug"]) as t:
    t.log("query", query)
    response = openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": query}],
    )
    t.log("response", response)
python
from opensmith import autopatch

autopatch()

Patch only selected backends:

from opensmith import autopatch

autopatch(only=["openai"])

Patch everything except selected backends:

from opensmith import autopatch

autopatch(exclude=["chromadb"])

Print trace results to the terminal as they complete:

from opensmith import set_console_mode, trace

set_console_mode(True)

@trace
def my_func():
    return "ok"

opensmith reads opensmith.json

from the current working directory on import:

Create a starter config:

opensmith init
{
  "db_path": "./my_traces.db",
  "console_mode": false,
  "autopatch": ["openai", "qdrant"]
}
opensmith ui

Open http://localhost:7823

. If the port is already in use, opensmith automatically tries the next available port. Use --no-auto-port

to disable this behavior.

Command	Description
opensmith init
Create a starter opensmith.json config.
opensmith ui
Start the local dashboard with automatic port selection.
opensmith traces --q rag --status err --tags production
List and filter traces in the terminal.
opensmith stats
Show aggregate trace, step, token, and cost statistics.
opensmith export
Export traces to JSON or CSV.
opensmith clear
Delete all locally stored traces after confirmation.

Backend	Package	Status
openai	openai	✅
anthropic	anthropic	✅
litellm	litellm	✅
qdrant	qdrant-client	✅
chromadb	chromadb	✅
pinecone	pinecone-client	✅

Traces are stored locally at ~/.opensmith/traces.db

unless overridden with opensmith.json

or set_default_db_path()

.

Use Postgres instead by installing opensmith[postgres]

and setting OPENSMITH_DB_URL

:

export OPENSMITH_DB_URL="postgresql://user:pass@localhost:5432/opensmith"

Install both Postgres and OpenTelemetry support with pip install "opensmith[all]"

.

Export traces with nested steps as JSON:

opensmith export --format json --output traces.json

Export a flat trace list as CSV:

opensmith export --format csv --output traces.csv

OpenTelemetry export is disabled by default. Install the optional dependencies and set OPENSMITH_OTEL_ENDPOINT

to enable OTLP HTTP export:

pip install "opensmith[otel]"
export OPENSMITH_OTEL_ENDPOINT="http://localhost:4318"

opensmith sends traces to /v1/traces

and metrics to /v1/metrics

under that endpoint.

MIT

source & further reading

github.com — original article