DEV Community

RESK
RESK

Posted on

ReskPoints: AI Agent Logging with Sampling, Masking and Multi-Export

#ai #opensource #python #devops

Links: PyPI: reskpoints | GitHub | resk.fr

The Problem

AI agents are making more autonomous decisions every day. But when they call a tool, modify state, or fetch external data, most teams rely on:

  • print statements
  • basic stdout
  • or nothing at all

That is un-auditable and dangerous in production. You need structured observability.

Enter ReskPoints

ReskPoints is an AI Agent Logger for Python that gives you:

  • configurable sampling per action type
  • automatic sensitive data masking
  • multi-export to Datadog, Prometheus, OpenTelemetry, console, and webhooks

Quick Start 

pip install reskpoints
Enter fullscreen mode Exit fullscreen mode 
from reskpoints import AgentLogger

logger = AgentLogger(
    sample_rate={"tool_call": 1.0, "heartbeat": 0.01},
    mask_fields=["api_key", "token", "password"],
    exporters=["console", "datadog"],
)

logger.log("tool_call", tool="search_web", query="LLM security")  # always logged
logger.log("heartbeat", status="ok")  # 1% sampled
Enter fullscreen mode Exit fullscreen mode

Key Features

  • Sampling control: 100% for critical actions, 1% for routine pings
  • Data masking: redact API keys, PII, tokens before export
  • Multi-export: ship to Datadog, Prometheus, OpenTelemetry, files, webhooks
  • Async batching: sub-ms overhead with zero blocking on your agent loop
  • Python >= 3.13, compatible with PyTorch >= 2.0.0

Why It Matters

If you deploy LLM agents in production without structured logging, you are flying blind. ReskPoints gives you the audit trail compliance teams require and the operational data engineers need.

Next Steps

Check the GitHub repo for docs and examples.

What do you use for AI agent observability?

Top comments (0)