trakr-monitor 0.1.0

AI agent cost and failure monitoring — Sentry for AI agents

trakr-monitor

Observability SDK for AI agent workflows. Automatically captures Anthropic and OpenAI LLM calls with cost, tokens, latency, and tool use — sent to your Trakr dashboard.

Coming soon on PyPI — not published yet. Use @trakr/monitor on npm for production today.

Install (when published)

pip install trakr-monitor

# With Anthropic auto-instrumentation
pip install trakr-monitor[anthropic]

# With OpenAI auto-instrumentation
pip install trakr-monitor[openai]

Quick start

import trakr_monitor as trakr
import anthropic

trakr.init()  # reads TRAKR_API_KEY

client = anthropic.Anthropic()

with trakr.start_run("daily-report") as run:
    client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=256,
        messages=[{"role": "user", "content": "Summarize logs"}],
    )

trakr.flush()  # blocking — call before sys.exit() in scripts

Decorator pattern

import trakr_monitor as trakr

trakr.init()

@trakr.track_agent(name="invoice-extractor")
def extract_invoice(document: str) -> dict:
    return do_extraction(document)

Async functions are supported:

@trakr.track_agent(name="async-processor")
async def process_async(data: str) -> str:
    return await call_llm(data)

Configuration

Option	Default	Description
api_key	TRAKR_API_KEY env	SDK API key (tk_live_...)
endpoint	https://app.trakr.run/api/ingest	Ingest URL override
batch_size	10	Events per upload batch
flush_interval_seconds	2.0	Max time between uploads
loop_detection_threshold	5	Consecutive same-tool calls before loop_detected
disabled	False	Disable all instrumentation
debug	False	Log SDK activity to stdout

Local development

import os
import trakr_monitor as trakr

trakr.init(
    api_key="tk_live_testkey00000000000000000000000000",
    endpoint=os.environ.get("TRAKR_INGEST_URL", "http://localhost:3000/api/ingest"),
)

API

trakr.init(...)
trakr.start_run(workflow_name, ...) -> Run  # context manager
trakr.track_agent(name, metadata=None)    # decorator
trakr.flush()                             # blocking flush
trakr.disable()

run.end(status="success" | "error", error_message=None)
run.set_metadata(key, value)

Tests

cd libs/sdk/python
pip install -e ".[dev]"
python -m pytest
python -m build

Documentation

• Getting started
• Python SDK
• Ingest API

Support

Questions or issues: [e-mail]

License

MIT