OpenInfra Logger is structured telemetry for Node, Python, Go and Rust — built from each runtime's standard library. Native batching, auto-redaction, and an offline AI analyzer for root-cause diagnosis.
npm i @jonathascordeiro20/openinfra-loggerfrom openinfra_logger import logimport openinfralogger "…/openinfra-logger/go"use openinfra_logger::Logger;Intercepts every event before it touches the wire. Passwords, API keys, JWTs, credit cards — replaced with [REDACTED] in the buffer. LGPD and GDPR-compliant out of the box, no rules to write.
Instead of HTTP-per-log, OIL holds events in a runtime-native buffer and ships them in lots of up to 100 every two seconds. No third-party transport. No backpressure surprises.
A CLI that reads your local error files and asks Claude to diagnose. Stack traces in, structured remediation out — your code never leaves your machine until you choose to send the prompt.
OIL formats and ships JSON identically regardless of language. Drop it into a polyglot stack — Datadog won't be able to tell which service wrote which event.
const { log } = require('@jonathascordeiro20/openinfra-logger'); log('order.placed', 'info', { userId: user.id, card: user.card // → [REDACTED] });
from openinfra_logger import log log("order.placed", "info", { "user_id": user.id, "card": user.card # → [REDACTED] })
import openinfralogger "github.com/jonathascordeiro20/openinfra-logger/go" openinfralogger.Log("order.placed", "info", map[string]interface{}{ "userId": user.ID, "card": user.Card // → [REDACTED] })
use openinfra_logger::{Logger, Config}; let logger = Logger::new(Config::default()); let mut md = HashMap::new(); md.insert("userId".to_string(), user.id.clone()); md.insert("card".to_string(), user.card.clone()); // → [REDACTED] logger.log("order.placed", "info", md);
{
"timestamp": "2026-05-15T14:08:22.041Z",
"status": "info",
"message": "order.placed",
"service": "checkout-api",
"runtime": "node.js",
"dd.trace_id": "a4f1c9…d3",
"userId": "u_8821",
"card": "[REDACTED]"
}
Switch destinations with a one-line config. OIL formats the wire payload natively for every major observability backend — no transformers, no exporters, no second sidecar.
status, dd.trace_id, dd.span_id@timestamp, log.leveltrace_id + span_id auto-linkJSON over HTTP pushstdout / file / remoteconfigurable remoteUrl + headersRun npm run analyze against any rolled error file. The CLI asks Claude to read the trace, identify the root cause, and propose a fix. Local files; explicit prompts; no telemetry phone-home.
$ npm run analyze ./logs/error-2026-05-15.jsonl 📄 Analyzing log file: ./logs/error-2026-05-15.jsonl... 🔍 Found 12 anomalies/errors. Preparing AI analysis... 🧠 Connecting to Anthropic Claude API... ◆ Root Cause Analysis db.timeout cascade originating in checkout.repo.findOrder(), triggered when order_items.user_id index is missing on the read replica (last migration: 2026-04-22). ◆ Suggested fix 1. add CREATE INDEX CONCURRENTLY on replica 2. raise findOrder timeout from 800ms → 2000ms 3. add trace_id to the retry budget key → wrote remediation to oil-analysis-a4f1.md $ █
The analyzer parses your rolled error logs without uploading them. Only the redacted, summarised prompt is sent to the model.
Uses your own ANTHROPIC_API_KEY. Pick haiku for speed, sonnet or opus for depth — or pipe the prompt to claude.ai manually.
The report is plain markdown / text — ready to attach to a PR, an incident doc, or a Slack thread.
[REDACTED] by the Shield. PII never reaches the model.One install, zero deps. Drop OIL into your Node, Python, Go, or Rust service and your telemetry is consistent on day one — across every team, every region, every backend.