lex-bedrock 0.2.1 → 0.2.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ee5a83a0ee7afd53d1d8f496bd8382332b1312939b3a49cbdf8c86b2269ff6ab
-  data.tar.gz: 467176f83f6f58de4a7640bea5f2fcd6b75be4e2a8d5a77d4fa8f3fe82854c2b
+  metadata.gz: 9c54111009d79155a7a9b85e0ec03a17b6b48800ce5dd641df7286637471114d
+  data.tar.gz: cb5f69ffe79c02ba38ba60c205b5215d7d8500851616ff5933c0e61ecf2f5b43
 SHA512:
-  metadata.gz: 106a987eb8c0d61d3c8fc1e409b0168d95277db70f98fba5ee8ebd0dbe9b74c3770837b052b6cb04b32aae8bfd639511de8b907dfaccee09ee72c563792ae051
-  data.tar.gz: d8c81a605d36fd3e6b572580ad4a092a1a7bf913e5edf51e5e26302facf20cd2248da4311fcbf1867fe6cfc35eea78bea61c71c4c7ca0ff43c91d99b920482f6
+  metadata.gz: 6cf4b797406f46d5b8b0f506913e429b90dc7aca0f17ead791476a1ab629f962c3c6495cc04f0cc16b9dc0be0b10704ac8546dbd164249aea378b30e9c1c290c
+  data.tar.gz: 1e886593369d42af3426a4bf30cda859642157d0f9199416b0ddb52c0ee5caf5621b6cbb0ad1cf2da324e112ade51c3cd2344a5ad3cecc5541dbb64a07166a4b

data/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+## [0.2.2] - 2026-04-06
+
+### Added
+- Broker soft consumer in Helpers::Client for AWS credential resolution via Identity::Broker (Phase 8 Wave 2)
+
 ## [0.2.1] - 2026-03-31
 
 ### Added

data/CLAUDE.md

@@ -6,29 +6,36 @@
 
 ## Purpose
 
-Legion Extension that connects LegionIO to AWS Bedrock for foundation model inference. Provides runners for the Converse API, raw InvokeModel, and model listing using the AWS SDK (not Faraday).
+Legion Extension that connects LegionIO to AWS Bedrock for foundation model inference. Provides runners for the Converse API, raw InvokeModel, model listing, token counting, and inference profile management using the AWS SDK (not Faraday).
 
 **GitHub**: https://github.com/LegionIO/lex-bedrock
 **License**: MIT
-**Version**: 0.1.2
-**Specs**: 38 examples
+**Version**: 0.2.2
+**Specs**: 120 examples (13 spec files)
 
 ## Architecture
 
 ```
 Legion::Extensions::Bedrock
 ├── Runners/
-│   ├── Converse       # create(model_id:, messages:, access_key_id:, secret_access_key:, ...)
-│   ├── Invoke         # invoke_model(model_id:, body:, access_key_id:, secret_access_key:, ...)
-│   └── Models         # list(access_key_id:, secret_access_key:, ...), get(model_id:, ...)
+│   ├── Converse       # create, create_stream, create_with_thinking
+│   ├── Invoke         # invoke_model
+│   ├── Models         # list, get
+│   ├── Tokens         # count_tokens
+│   └── Profiles       # list_inference_profiles, get_inference_profile, resolve_profile_id
 ├── Helpers/
-│   └── Client         # AWS SDK client factories (module, not Faraday)
+│   ├── Client         # AWS SDK client factories (module, not Faraday)
+│   ├── Credentials    # credential resolution helpers
+│   ├── Errors         # error handling + with_retry
+│   ├── ModelRegistry  # canonical model name -> Bedrock cross-region profile ID mapping
+│   ├── Thinking       # extended thinking field builders
+│   └── Usage          # usage normalization
 └── Client             # Standalone client class (includes all runners, holds @config)
 ```
 
 `Helpers::Client` is a **module** with two factory methods:
-- `bedrock_runtime_client(...)` — builds `Aws::BedrockRuntime::Client` (used by Converse and Invoke runners)
-- `bedrock_client(...)` — builds `Aws::Bedrock::Client` (used by Models runner)
+- `bedrock_runtime_client(...)` — builds `Aws::BedrockRuntime::Client` (used by Converse, Invoke, Tokens runners)
+- `bedrock_client(...)` — builds `Aws::Bedrock::Client` (used by Models, Profiles runners)
 
 `DEFAULT_REGION` is `'us-east-2'`. All credential kwargs (`access_key_id:`, `secret_access_key:`, `region:`, `session_token:`) are passed through to the AWS SDK constructors. `session_token:` is optional and omitted if nil.
 
@@ -36,25 +43,32 @@ Legion::Extensions::Bedrock
 
 - This is the only extension in extensions-ai that uses the AWS SDK instead of Faraday. There is no HTTP connection object — SDK clients handle transport.
 - `Invoke#invoke_model` uses `::JSON.dump` and `::JSON.parse` (stdlib, explicit `::` prefix) to avoid namespace collision with `Legion::JSON`.
-- `Converse#create` returns `{ result: response.output, usage: response.usage, stop_reason: response.stop_reason }` — three keys, unlike other runners which return only `{ result: ... }`.
-- `Models#list` returns `{ models: response.model_summaries }` and `Models#get` returns `{ model: response.model_details }`.
+- `Converse#create` returns `{ result: response.output, usage: response.usage, stop_reason: response.stop_reason }`.
+- `Converse#create_stream` accepts a block and yields `{ type: :delta, text: }`, `{ type: :stop, stop_reason: }`, `{ type: :usage, usage: }` events. Returns `{ result: accumulated_text, usage:, stop_reason: }`.
+- `Converse#create_with_thinking` wraps `create` with `Helpers::Thinking.build_thinking_fields`. Supports `budget_tokens:`, `adaptive:`, `extra_betas:`.
+- `Tokens#count_tokens` uses `client.count_tokens(...)` SDK call; returns `{ input_token_count: response.input_tokens }`.
+- `Profiles` runner lists, gets, and resolves AWS cross-region inference profiles. `resolve_profile_id` finds the profile ARN for a canonical model ID.
+- `Helpers::ModelRegistry` maps canonical model names (e.g. `'claude-sonnet-4-6'`) to Bedrock cross-region profile IDs (e.g. `'us.anthropic.claude-sonnet-4-6'`). Use `ModelRegistry.resolve(model_id)` before passing to SDK calls.
+- `Helpers::Thinking` constants: `THINKING_BETA = 'interleaved-thinking-2025-05-14'`, `CONTEXT_1M_BETA = 'context-1m-2025-08-07'`, `TOOL_SEARCH_BETA = 'tool-search-tool-2025-10-19'`. Supports both `adaptive` and `enabled` thinking modes.
 - `include Legion::Extensions::Helpers::Lex` is guarded with `const_defined?` pattern.
 
 ## Dependencies
 
 | Gem | Purpose |
 |-----|---------|
-| `aws-sdk-bedrock` | Bedrock management API (model listing) |
-| `aws-sdk-bedrockruntime` | Bedrock runtime API (inference) |
+| `aws-sdk-bedrock` | Bedrock management API (model listing, inference profiles) |
+| `aws-sdk-bedrockruntime` | Bedrock runtime API (inference, streaming, token counting) |
+| `legion-cache`, `legion-crypt`, `legion-data`, `legion-json`, `legion-logging`, `legion-settings`, `legion-transport` | LegionIO core |
 
 ## Testing
 
 ```bash
 bundle install
-bundle exec rspec        # 38 examples
+bundle exec rspec        # 120 examples
 bundle exec rubocop
 ```
 
 ---
 
 **Maintained By**: Matthew Iverson (@Esity)
+**Last Updated**: 2026-04-06

data/README.md

@@ -4,7 +4,7 @@ Legion Extension that connects LegionIO to AWS Bedrock for foundation model infe
 
 ## Purpose
 
-Wraps the AWS Bedrock SDK as named runners consumable by any LegionIO task chain. Provides the Converse API (multi-turn chat), raw InvokeModel (direct model invocation), and foundation model listing. Uses the AWS SDK directly — no Faraday. Use this extension when you need direct access to the Bedrock API surface within the LEX runner/actor lifecycle. For simple chat/embed workflows, consider `legion-llm` instead.
+Wraps the AWS Bedrock SDK as named runners consumable by any LegionIO task chain. Provides the Converse API (multi-turn chat with streaming and extended thinking), raw InvokeModel (direct model invocation), foundation model listing, token counting, and inference profile management. Uses the AWS SDK directly — no Faraday. Use this extension when you need direct access to the Bedrock API surface within the LEX runner/actor lifecycle. For simple chat/embed workflows, consider `legion-llm` instead.
 
 ## Installation
 
@@ -40,17 +40,46 @@ result = client.list
 
 # Converse API
 result = client.create(
-  model_id: 'anthropic.claude-3-5-sonnet-20241022-v2:0',
+  model_id: 'us.anthropic.claude-sonnet-4-6',
   messages: [{ role: 'user', content: [{ text: 'Hello' }] }]
 )
 # => { result: ..., usage: ..., stop_reason: 'end_turn' }
 
+# Streaming Converse
+client.create_stream(
+  model_id: 'us.anthropic.claude-sonnet-4-6',
+  messages: [{ role: 'user', content: [{ text: 'Tell me a story.' }] }]
+) do |event|
+  case event[:type]
+  when :delta then print event[:text]
+  when :stop  then puts "\nStop: #{event[:stop_reason]}"
+  when :usage then puts event[:usage].inspect
+  end
+end
+
+# Extended thinking
+result = client.create_with_thinking(
+  model_id: 'us.anthropic.claude-sonnet-4-6',
+  messages: [{ role: 'user', content: [{ text: 'Solve this step by step.' }] }],
+  budget_tokens: 8192
+)
+
+# Token counting
+result = client.count_tokens(
+  model_id: 'us.anthropic.claude-sonnet-4-6',
+  messages: [{ role: 'user', content: [{ text: 'Hello' }] }]
+)
+# => { input_token_count: 10 }
+
 # Raw InvokeModel
 result = client.invoke_model(
   model_id: 'anthropic.claude-3-5-sonnet-20241022-v2:0',
   body: { prompt: 'Hello', max_tokens: 256 }
 )
-# => { result: { ... }, content_type: 'application/json' }
+
+# Resolve model ID via ModelRegistry
+resolved = Legion::Extensions::Bedrock::Helpers::ModelRegistry.resolve('claude-sonnet-4-6')
+# => 'us.anthropic.claude-sonnet-4-6'
 ```
 
 ### Runner Modules
@@ -59,15 +88,45 @@ result = client.invoke_model(
 include Legion::Extensions::Bedrock::Runners::Models
 include Legion::Extensions::Bedrock::Runners::Converse
 include Legion::Extensions::Bedrock::Runners::Invoke
+include Legion::Extensions::Bedrock::Runners::Tokens
+include Legion::Extensions::Bedrock::Runners::Profiles
 ```
 
 ## API Coverage
 
 | Runner | Methods |
 |--------|---------|
-| `Converse` | `create` |
+| `Converse` | `create`, `create_stream`, `create_with_thinking` |
 | `Invoke` | `invoke_model` |
 | `Models` | `list`, `get` |
+| `Tokens` | `count_tokens` |
+| `Profiles` | `list_inference_profiles`, `get_inference_profile`, `resolve_profile_id` |
+
+## Model Registry
+
+`Helpers::ModelRegistry` maps canonical model names to Bedrock cross-region inference profile IDs:
+
+```ruby
+Legion::Extensions::Bedrock::Helpers::ModelRegistry.resolve('claude-sonnet-4-6')
+# => 'us.anthropic.claude-sonnet-4-6'
+
+Legion::Extensions::Bedrock::Helpers::ModelRegistry.all
+# => { 'claude-3-5-haiku-20241022' => 'us.anthropic.claude-3-5-haiku-20241022-v1:0', ... }
+```
+
+## Extended Thinking
+
+`Helpers::Thinking` provides field builders for Anthropic extended thinking on Bedrock:
+
+```ruby
+# Fixed budget
+fields = Legion::Extensions::Bedrock::Helpers::Thinking.build_thinking_fields(budget_tokens: 8192)
+
+# Adaptive thinking
+fields = Legion::Extensions::Bedrock::Helpers::Thinking.build_thinking_fields(adaptive: true)
+```
+
+Constants: `THINKING_BETA = 'interleaved-thinking-2025-05-14'`, `CONTEXT_1M_BETA = 'context-1m-2025-08-07'`, `TOOL_SEARCH_BETA = 'tool-search-tool-2025-10-19'`.
 
 ## Related
 
@@ -75,6 +134,10 @@ include Legion::Extensions::Bedrock::Runners::Invoke
 - `legion-llm` — High-level LLM interface including Bedrock via ruby_llm
 - `extensions-ai/CLAUDE.md` — Architecture patterns shared across all AI extensions
 
+## Version
+
+0.2.1
+
 ## License
 
 MIT

data/lib/legion/extensions/bedrock/helpers/client.rb

@@ -15,21 +15,17 @@ module Legion
           def bedrock_runtime_client(access_key_id: nil, secret_access_key: nil,
                                      region: DEFAULT_REGION, session_token: nil,
                                      credentials: nil, **)
-            Aws::BedrockRuntime::Client.new(
-              region:,
-              credentials: credentials || build_credentials(access_key_id:, secret_access_key:,
-                                                            session_token:)
-            )
+            credentials ||= build_credentials(access_key_id:, secret_access_key:, session_token:)
+            credentials ||= resolve_broker_credentials
+            Aws::BedrockRuntime::Client.new(region:, credentials:)
           end
 
           def bedrock_client(access_key_id: nil, secret_access_key: nil,
                              region: DEFAULT_REGION, session_token: nil,
                              credentials: nil, **)
-            Aws::Bedrock::Client.new(
-              region:,
-              credentials: credentials || build_credentials(access_key_id:, secret_access_key:,
-                                                            session_token:)
-            )
+            credentials ||= build_credentials(access_key_id:, secret_access_key:, session_token:)
+            credentials ||= resolve_broker_credentials
+            Aws::Bedrock::Client.new(region:, credentials:)
           end
 
           def region_for_model(model_id:, region: nil)
@@ -48,6 +44,18 @@ module Legion
             settings_region || DEFAULT_REGION
           end
 
+          def resolve_broker_credentials
+            return nil unless defined?(Legion::Identity::Broker)
+
+            renewer = Legion::Identity::Broker.renewer_for(:aws)
+            return nil unless renewer&.provider.respond_to?(:current_credentials)
+
+            renewer.provider.current_credentials
+          rescue StandardError => e
+            log.warn("resolve_broker_credentials failed: #{e.message}")
+            nil
+          end
+
           def build_credentials(access_key_id:, secret_access_key:, session_token:)
             return nil if access_key_id.nil?
 
@@ -58,7 +66,7 @@ module Legion
             end
           end
 
-          private_class_method :build_credentials
+          private_class_method :build_credentials, :resolve_broker_credentials
         end
       end
     end

data/lib/legion/extensions/bedrock/version.rb

@@ -3,7 +3,7 @@
 module Legion
   module Extensions
     module Bedrock
-      VERSION = '0.2.1'
+      VERSION = '0.2.2'
     end
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-bedrock
 version: !ruby/object:Gem::Version
-  version: 0.2.1
+  version: 0.2.2
 platform: ruby
 authors:
 - Esity