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

#observability #ai #opensource #python

ReskPoints makes every agent action observable. Console, Datadog, Prometheus, OpenTelemetry — one install, all exporters.

Links:

GitHub: https://github.com/Resk-Security/ReskPoints
PyPI: https://pypi.org/project/reskpoints
Website: https://resk.fr

The Problem

A typical AI agent orchestrator makes dozens of tool calls per user request. Each call has request parameters a response latency and a success or failure state. When something goes wrong you need to replay what the agent was thinking and doing.

Standard Python logging gives you raw text. Datadog APM gives you traces but not the agent intent layer. You need a purpose-built logger that captures the agent loop not just the HTTP calls.

The Solution

ReskPoints is a Python library that hooks into your agent loop and records every action with context. Here is the core API:

from reskpoints import AgentLogger

logger = AgentLogger(
    service="my-agent",
    sampling_rate=0.1,   # keep 10 percent of actions
    exporters=[
        "console",
        "datadog",
        "prometheus"
    ],
    masks=[
        "api_key",
        "user.email"
    ]
)

# Inside your agent loop
logger.log_action(
    action="tool_call",
    tool="search_docs",
    input={"query": user_input},
    output=result,
    duration_ms=340,
    token_count=1200
)

Key Features

Sampling — keep 10 percent of actions and 100 percent of errors. Saves costs on high-volume agents.
Masking — regex-based field protection. Sensitive data never leaves the agent process.
Multi-export — write to Console, Datadog, Prometheus, OpenTelemetry, webhooks, or local files. Swap without changing agent code.
Zero-overhead when idle — no background threads or polling loops.

The library works with Python 3.13+ and has zero required dependencies beyond the exporters you use.