sdk-python

Sentro Python SDK

Error tracking and agent observability for AI agents. Zero dependencies.

Sentro is an open-source Sentry alternative built specifically for AI agents. This SDK gives you full observability into your agent runs — every step, tool call, LLM call, token count, and cost.

Install

pip install sentro-sdk

Quick Start

Error tracking

from sentro import Sentro

sentro = Sentro(dsn="http://token@localhost:3000/api/ingest/proj_1")

try:
    do_something()
except Exception as e:
    sentro.capture_exception(e)

Agent observability

from sentro import Sentro

sentro = Sentro(dsn="http://token@localhost:3000/api/ingest/proj_1")

with sentro.trace("order-processor", goal="Process refund #456") as run:
    with run.trace("Looking up order") as step:
        # Track tool calls
        with step.trace_tool_call("db.query", input={"sql": "SELECT 1"}) as tool:
            result = db.query("SELECT 1")
            tool.set_result(result)

        # Track LLM calls
        llm = step.llm_call(model="claude-sonnet-4-6")
        response = call_llm("Approve refund?")
        llm.end(prompt_tokens=150, completion_tokens=20, cost=0.001)
# Auto-ends on exit, auto-errors on exception

Async support

async with sentro.trace("async-agent", goal="Async task") as run:
    async with run.trace("Step 1") as step:
        result = await some_async_work()

Features

• Zero dependencies — standard library only, works in any Python 3.10+ environment
• Context managers — with blocks for automatic lifecycle management
• Async support — async with for async frameworks
• Auto error capture — exceptions automatically mark runs as failed
• Batched transport — events are batched and sent efficiently
• Run tracing — full agent execution timeline
• Step replay — ordered reasoning chain with tool and LLM call details
• Cost tracking — token counts and cost per LLM call

Configuration

sentro = Sentro(
    dsn="http://token@localhost:3000/api/ingest/proj_1",
    default_tags={"env": "production", "version": "1.0.0"},
)

Security Considerations

• The capture_prompts option is disabled by default to avoid storing LLM prompt/response bodies that may contain PII
• DSN tokens are API keys — treat them like passwords, don't commit them to source control
• Tool call inputs/outputs are stored as-is — avoid passing sensitive data (API keys, passwords) through traced tool calls

Framework Integrations

LangChain

from sentro import Sentro
from sentro.integrations.langchain import SentroMiddleware

sentro = Sentro(dsn="YOUR_DSN")

agent = create_agent(
    model="gpt-4o",
    tools=[...],
    middleware=[SentroMiddleware(sentro)],
)

Requires pip install langchain langgraph.

CrewAI

from sentro import Sentro
from sentro.integrations.crewai import SentroCrewListener

sentro = Sentro(dsn="YOUR_DSN")
SentroCrewListener(sentro)  # auto-registers — just kick off your crew

Requires pip install crewai.

Links

• GitHub: github.com/yzzztech/sentro
• Docs: Available at /docs when running Sentro
• TypeScript SDK: npm install @sentro/sdk

License

MIT