tracekit 0.2.2 → 0.2.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7a2a8b53b2c100727c3161bd201a6e310135d603c7ac7ab1e8370417a4ff2342
-  data.tar.gz: 710eb5980307b7a4e82a698a86ff1dea131af93a24b0785f9d88c07a1cb4288a
+  metadata.gz: 9133bb6f1112ae31a4732cf52a12d209fa908e6f0acb5af92d3bf566ecc99ff5
+  data.tar.gz: dd8eeb7476eef7bc591833194b8805c9f0c74f8c1b8e389a4b34834ceea5c201
 SHA512:
-  metadata.gz: 7548453f27b8b0781cda582feb75c25479e5abb5d2cf670c969055402dd6cea6d15582df725038112e35f277045d13baa6b82645295aa8b8a348349b4a0a0d17
-  data.tar.gz: 9e103dc9c20d0a00182bc403df79cb7fe8043bc58adf68a63a76996ed0952183a5251b94f6c35fdc47957e71e1934ab4365d630a1dd2268d1c85afec71668727
+  metadata.gz: 10f4ff9f1123866d67485d7b02a1224a22abc9f3e05814064d8057103172c7a13a7a527a8626b612a622df3feed3f9ec0ddb632550d669a1a62f46a18354c86f
+  data.tar.gz: 3b9c8a7b79eb316ef4322b1b54deba4bc9e6b3538ff2c3fdefb6b4cfff5b394e6003d4dcbe225d837f2da5fa6f246044e1c6052d539b6e7a1063317c71b639a8

data/CHANGELOG.md

@@ -5,6 +5,22 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.3] - 2026-03-21
+
+### Added
+- LLM auto-instrumentation for OpenAI and Anthropic APIs via Module#prepend
+- Streaming support for both OpenAI (SSE chunks) and Anthropic (SSE events) chat completions
+- Automatic capture of gen_ai.* semantic convention attributes (model, provider, tokens, cost, latency, finish_reason)
+- Content capture option for request/response messages (TRACEKIT_LLM_CAPTURE_CONTENT env var)
+- Tool call detection and instrumentation for function calling
+- PII scrubbing for captured LLM content
+- Provider auto-detection via LoadError handling
+- StreamWrapper for OpenAI and AnthropicStreamWrapper for Anthropic streaming responses
+- Anthropic cache token tracking (cache_creation_input_tokens, cache_read_input_tokens)
+
+### Changed
+- SDK init auto-detects and patches OpenAI::Client#chat and Anthropic::Client::Messages#create when gems are present
+
 ## [0.1.0] - 2024-02-04
 
 ### Added
@@ -136,3 +152,4 @@ This is the first production-ready release of the TraceKit Ruby SDK. It provides
 ---
 
 [0.1.0]: https://github.com/Tracekit-Dev/ruby-sdk/releases/tag/v0.1.0
+[0.2.3]: https://github.com/Tracekit-Dev/ruby-sdk/releases/tag/v0.2.3

data/README.md

@@ -24,6 +24,7 @@ TraceKit Ruby SDK provides production-ready distributed tracing, metrics, and co
 - **Code Monitoring**: Live production debugging with non-breaking snapshots
 - **Security Scanning**: Automatic detection of sensitive data (PII, credentials)
 - **Local UI Auto-Detection**: Automatically sends traces to local TraceKit UI
+- **LLM Auto-Instrumentation**: Zero-config tracing of OpenAI and Anthropic API calls via Module#prepend
 - **Rails Auto-Configuration**: Zero-configuration setup via Railtie
 - **Rack Middleware**: Automatic request instrumentation for any Rack application
 - **Thread-Safe Metrics**: Concurrent metric collection with automatic buffering
@@ -377,6 +378,100 @@ sdk.capture_snapshot("process-data", { batch_size: 100 })
 - The SDK automatically retries after the cooldown period
 - Thread-safe via `Mutex` — safe for multi-threaded Ruby applications (Puma, Sidekiq)
 
+## LLM Instrumentation
+
+TraceKit automatically instruments OpenAI and Anthropic API calls when the gems are present. No manual setup required — the SDK patches clients at init via `Module#prepend`.
+
+### Supported Gems
+
+- **[ruby-openai](https://github.com/alexrudall/ruby-openai)** (~> 7.0) — `OpenAI::Client#chat`
+- **[anthropic](https://github.com/alexrudall/anthropic)** (~> 0.3) — `Anthropic::Client#messages`
+
+### Usage
+
+```ruby
+# Just use the gems normally — TraceKit instruments automatically
+
+# OpenAI
+client = OpenAI::Client.new(access_token: ENV["OPENAI_API_KEY"])
+response = client.chat(parameters: {
+  model: "gpt-4o-mini",
+  messages: [{ role: "user", content: "Hello!" }],
+  max_tokens: 100
+})
+
+# Anthropic
+client = Anthropic::Client.new(access_token: ENV["ANTHROPIC_API_KEY"])
+response = client.messages(parameters: {
+  model: "claude-sonnet-4-20250514",
+  max_tokens: 100,
+  messages: [{ role: "user", content: "Hello!" }]
+})
+```
+
+### Streaming
+
+Both streaming and non-streaming calls are instrumented:
+
+```ruby
+# OpenAI streaming
+client.chat(parameters: {
+  model: "gpt-4o-mini",
+  messages: [{ role: "user", content: "Tell me a story" }],
+  stream: proc { |chunk, _bytesize|
+    print chunk.dig("choices", 0, "delta", "content")
+  }
+})
+
+# Anthropic streaming
+client.messages(parameters: {
+  model: "claude-sonnet-4-20250514",
+  max_tokens: 200,
+  messages: [{ role: "user", content: "Tell me a story" }],
+  stream: proc { |event|
+    if event["type"] == "content_block_delta"
+      print event.dig("delta", "text")
+    end
+  }
+})
+```
+
+### Captured Attributes
+
+Each LLM call creates a span with [GenAI semantic convention](https://opentelemetry.io/docs/specs/semconv/gen-ai/) attributes:
+
+| Attribute | Description |
+|-----------|-------------|
+| `gen_ai.system` | `openai` or `anthropic` |
+| `gen_ai.request.model` | Model name (e.g., `gpt-4o-mini`) |
+| `gen_ai.request.max_tokens` | Max tokens requested |
+| `gen_ai.response.model` | Model used in response |
+| `gen_ai.response.id` | Response ID |
+| `gen_ai.response.finish_reason` | `stop`, `end_turn`, etc. |
+| `gen_ai.usage.input_tokens` | Prompt tokens used |
+| `gen_ai.usage.output_tokens` | Completion tokens used |
+
+### Content Capture
+
+Input/output content capture is **disabled by default** for privacy. Enable it with:
+
+```bash
+TRACEKIT_LLM_CAPTURE_CONTENT=true
+```
+
+### Configuration
+
+LLM instrumentation is enabled by default when OpenAI or Anthropic gems are detected. To disable:
+
+```ruby
+Tracekit.configure do |config|
+  config.llm = { enabled: false }          # Disable all LLM instrumentation
+  config.llm = { openai: false }           # Disable OpenAI only
+  config.llm = { anthropic: false }        # Disable Anthropic only
+  config.llm = { capture_content: true }   # Enable content capture via config
+end
+```
+
 ## Distributed Tracing
 
 The SDK automatically:
@@ -452,6 +547,10 @@ ruby-sdk/
 │   │   ├── sdk.rb                 # Main SDK class
 │   │   ├── railtie.rb             # Rails auto-configuration
 │   │   ├── middleware.rb          # Rack middleware
+│   │   ├── llm/                   # LLM auto-instrumentation
+│   │   │   ├── common.rb          # Shared helpers, PII scrubbing
+│   │   │   ├── openai_instrumentation.rb   # OpenAI Module#prepend
+│   │   │   └── anthropic_instrumentation.rb # Anthropic Module#prepend
 │   │   ├── metrics/               # Metrics implementation
 │   │   ├── security/              # Security scanning
 │   │   └── snapshots/             # Code monitoring
@@ -523,6 +622,7 @@ bundle exec rails server -p 5002
 - `GET /api/call-go` - Call Go test service
 - `GET /api/call-node` - Call Node test service
 - `GET /api/call-all` - Call all test services
+- `GET /api/llm` - LLM instrumentation test (OpenAI + Anthropic, streaming + non-streaming)
 
 See [ruby-test/README.md](ruby-test/README.md) for details.
 
@@ -597,4 +697,4 @@ Built on [OpenTelemetry](https://opentelemetry.io/) - the industry standard for
 ---
 
 **Repository**: git@github.com:Tracekit-Dev/ruby-sdk.git
-**Version**: v0.2.0
+**Version**: v0.2.3

data/lib/tracekit/config.rb

@@ -6,7 +6,8 @@ module Tracekit
   class Config
     attr_reader :api_key, :service_name, :endpoint, :use_ssl, :environment,
                 :service_version, :enable_code_monitoring,
-                :code_monitoring_poll_interval, :local_ui_port, :sampling_rate
+                :code_monitoring_poll_interval, :local_ui_port, :sampling_rate,
+                :llm
 
     def initialize(builder)
       @api_key = builder.api_key
@@ -19,6 +20,7 @@ module Tracekit
       @code_monitoring_poll_interval = builder.code_monitoring_poll_interval || 30
       @local_ui_port = builder.local_ui_port || 9999
       @sampling_rate = builder.sampling_rate || 1.0
+      @llm = (builder.llm || {}).freeze
 
       validate!
       freeze # Make configuration immutable
@@ -35,7 +37,8 @@ module Tracekit
     class Builder
       attr_accessor :api_key, :service_name, :endpoint, :use_ssl, :environment,
                     :service_version, :enable_code_monitoring,
-                    :code_monitoring_poll_interval, :local_ui_port, :sampling_rate
+                    :code_monitoring_poll_interval, :local_ui_port, :sampling_rate,
+                    :llm
 
       def initialize
         # Set defaults in builder
@@ -47,6 +50,7 @@ module Tracekit
         @code_monitoring_poll_interval = 30
         @local_ui_port = 9999
         @sampling_rate = 1.0
+        @llm = { enabled: true, openai: true, anthropic: true, capture_content: false }
       end
     end