Strawberry Ruby SDK

Non-blocking analytics, error tracking, and logging for Ruby servers.

Stdlib only. No runtime dependencies. Ruby 3.0+.

Install

Add to your Gemfile:

gem "strawberry-sdk", git: "https://github.com/steventruong/strawberry-ruby.git", tag: "v1.0.0"

Then bundle install.

Usage

require "strawberry"

Strawberry.init(
  api_key: ENV.fetch("STRAWBERRY_API_KEY"),
  host: "https://straw.berryagents.com",
  release_version: ENV["GIT_SHA"],
  environment: "production"
)

Strawberry.capture("order_placed", { amount: 42.50, currency: "USD" }, distinct_id: "user_123")
Strawberry.identify("user_123", { email: "jane@example.com", plan: "pro" })

begin
  do_work
rescue => e
  Strawberry.capture_error(e, { endpoint: "/checkout" })
end

Strawberry.log(:info, "cache miss", category: "cache", attrs: { key: "u_123" })

Strawberry.llm_call(
  provider: "openai",
  model: "gpt-4o",
  prompt_tokens: 120,
  completion_tokens: 80,
  latency_ms: 412,
  cost_usd: 0.002,
  status: "success"
)

Strawberry.flush    # force drain
Strawberry.shutdown # stop worker (called automatically at_exit)

Rack middleware

# config.ru
require "strawberry"

Strawberry.init(api_key: ENV.fetch("STRAWBERRY_API_KEY"), host: "https://app.getstrawberry.io")

use Strawberry.middleware
run MyApp

Architecture

• Per-channel bounded queues (events, errors, logs, identify, llm) with drop-oldest overflow
• Per-endpoint circuit breaker (CLOSED / OPEN / HALF_OPEN)
• Jittered decorrelated backoff on retry
• Always-on PII redactor (emails, phones, Luhn-valid cards, JWTs, API-key prefixes, field-name denylist)
• Single background worker thread, at_exit flush
• Injectable transport + clock for tests

Diagnostics

Strawberry.diagnostics
# => {
#   enabled: true,
#   host: "https://straw.berryagents.com",
#   queue_sizes: { events: 0, errors: 0, logs: 0, identify: 0, llm: 0 },
#   dropped: { events: 0, errors: 0, logs: 0, identify: 0, llm: 0 },
#   circuits: { ingest: :closed, errors: :closed, logs: :closed }
# }

Guarantees

• Public calls never raise. Observability never breaks the caller.
• Transport is asynchronous. capture returns immediately after enqueue.
• Overflow drops oldest items, so the newest signal always wins.
• On transport failure items are requeued (bounded by cap) and retried with jittered backoff.