strawberry-sdk (Node.js)

Node.js SDK for Strawberry analytics, error tracking, logs, and LLM call observability. Stdlib-only (no dependencies), Node 18+.

Features:

• Per-channel bounded buffers (events, errors, logs, identify, llm) with drop-oldest on overflow
• Per-endpoint circuit breaker (CLOSED / OPEN / HALF_OPEN)
• Jittered decorrelated backoff for retries
• Always-on PII redaction (email, phone, Luhn-valid cards, JWTs, API keys, field-name denylist)
• Injectable transport and clock for tests, plus dryRun
• Graceful shutdown with timeout and a process.on('exit') hook
• Diagnostics accessor for queue depth, drops, retries, failures, circuit states, redactions

Install

npm install github:steventruong/strawberry-node#v1.0.0

Requires Node 18+.

Quick start

const strawberry = require('strawberry-sdk');

strawberry.init('berry_xxxxxxxx', {
  host: 'https://straw.berryagents.com',
  releaseVersion: 'v1.2.3',
  environment: 'production',
});

strawberry.capture('user_signed_up', { plan: 'pro' }, 'user_123');
strawberry.identify('user_123', { email: 'a@b.com', plan: 'pro' });
strawberry.log('info', 'sync finished', 'worker', { items: 42 });
strawberry.llmCall({
  provider: 'openai',
  model: 'gpt-4o',
  promptTokens: 120,
  completionTokens: 48,
  latencyMs: 812,
  costUsd: 0.003,
});

try { risky(); } catch (e) { strawberry.captureError(e, { op: 'risky' }); }

process.on('beforeExit', async () => { await strawberry.shutdown(); });

Express middleware

const app = require('express')();
app.use(strawberry.middleware());

Emits $http_request on every finished response and $http_error on 5xx or caught errors.

Public API

Function	Notes
init(apiKey, opts?)	{ host, releaseVersion, environment, batchSize, flushInterval, transport, clock, redactor, dryRun }
capture(event, props?, distinctId?)	Non-blocking.
identify(distinctId, props?)	Non-blocking.
captureError(err, context?)	Accepts Error or string.
log(level, message, category?, attrs?)	Level is one of error, warn, info, debug.
llmCall({ provider, model, promptTokens?, completionTokens?, latencyMs?, costUsd?, status?, streaming?, distinctId?, errorMessage? })
flush()	Returns Promise<void>.
shutdown()	Returns Promise<void>.
isEnabled()
diagnostics()	Snapshot of queue depth, drops, retries, failures, circuit states, redactions.
middleware()	Express / Connect compatible.

Wire protocol

• POST /api/v1/ingest — batched events + identify + LLM calls
• POST /api/v1/errors/ingest — one error per request
• POST /api/v1/logs — one structured log per request
• Auth via Authorization: Bearer <apiKey> plus api_key body fallback

Design notes

• Buffers never grow unbounded. Overflow drops the oldest entry and increments a dropped counter visible in diagnostics().
• The circuit breaker is per-endpoint — a flaky errors endpoint doesn't starve analytics.
• Retries use decorrelated jittered backoff to avoid synchronized stampedes under contention.
• The PII redactor runs on every value before it enters a buffer, so persisted queues never contain unredacted data.
• dryRun: true short-circuits all HTTP, useful for local development and tests.