legion-llm 0.9.32 → 0.9.33

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 715bd8c0918939545eda0cf832d81aa23e69c807ed6055fdb4d95f4177c99449
-  data.tar.gz: 12e42d3d2fdc02c4ca7764a264af90631141952982d9726c02d0e9ea7de87a92
+  metadata.gz: 768503fd466d914e0cc3f4bb713ac0a710b392ebc0098eab5d6c2def05f5f9f5
+  data.tar.gz: 9b0823a06222164a83123d0b895c701da3415aee51bca8b8ab303e6f1b72a532
 SHA512:
-  metadata.gz: 6f2fd1a0ea8b18ed222f2713adb4f4a48ce57e90d5a3ac2242f7ae648ed297b6375d1143c0e29777e4c5ade60e479e09997c2af04cf0e1d3b81225ef3a14276f
-  data.tar.gz: 41d0daa21a98518c4192881bd3231ca1e308b6231d06e22b6102531feee4f6aef1ced50d9e43b1f65b7e8f344e95406ebb682bf445d356011aafd6d4ce241a37
+  metadata.gz: 8b85a0b079619ed107b79044b6cfaa29f58a11881cd67a883cc2d107efebbf19240241d892daf983abd89876391fb3152434da27ce7e3a043f9170bde75d747e
+  data.tar.gz: 3c64a0f8744c10882cf02510480dd9757c68053454dabfd639340b16561f8640011fba2d4cb120d32c4ee532ac497db10167c516656562f722aa415f2800e49a

data/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Legion LLM Changelog
 
+## [0.9.33] - 2026-05-22
+
+### Added
+- API: OpenAI Responses API endpoint (`POST /v1/responses` and `POST /api/llm/inference/v1/responses`) for Codex CLI compatibility
+- API: Supports streaming (SSE with `response.*` events), tool declarations, and `instructions` parameter
+
 ## [0.9.31] - 2026-05-18
 
 ### Added

data/lib/legion/llm/api/openai/responses.rb

@@ -0,0 +1,271 @@
+# frozen_string_literal: true
+
+require 'securerandom'
+require 'legion/logging/helper'
+require 'legion/llm/types'
+
+module Legion
+  module LLM
+    module API
+      module OpenAI
+        module Responses
+          extend Legion::Logging::Helper
+
+          def self.registered(app)
+            log.debug('[llm][api][openai][responses] registering POST /v1/responses + /api/llm/inference/v1/responses')
+
+            handler = build_handler
+
+            app.post('/v1/responses') { instance_exec(&handler) }
+            app.post('/api/llm/inference/v1/responses') { instance_exec(&handler) }
+
+            log.debug('[llm][api][openai][responses] routes registered')
+          end
+
+          def self.build_handler # rubocop:disable Metrics/MethodLength
+            proc do # rubocop:disable Metrics/BlockLength
+              require_llm!
+              body = parse_request_body
+              request_id = "resp_#{SecureRandom.hex(16)}"
+
+              input = body[:input]
+              messages = case input
+                         when Array
+                           Responses.normalize_input_array(input)
+                         when String
+                           [{ role: 'user', content: input }]
+                         else
+                           halt 400, { 'Content-Type' => 'application/json' },
+                                Legion::JSON.dump({ error: { message: 'input is required (string or array)',
+                                                             type: 'invalid_request_error', code: nil } })
+                         end
+
+              messages = [{ role: 'system', content: body[:instructions].to_s }] + messages if body[:instructions]
+
+              model = body[:model] || Legion::LLM::Settings.value(:default_model) || 'default'
+              streaming = body[:stream] == true
+
+              tool_declarations = Responses.build_tool_declarations(body[:tools])
+
+              log.info(
+                "[llm][api][openai][responses] action=accepted request_id=#{request_id} " \
+                "model=#{model} stream=#{streaming} tools=#{tool_declarations.size}"
+              )
+
+              effective_caller = build_server_caller(source: 'openai_responses', path: request.path, env: env)
+
+              require 'legion/llm/inference/request' unless defined?(Legion::LLM::Inference::Request)
+              require 'legion/llm/inference/executor' unless defined?(Legion::LLM::Inference::Executor)
+
+              inference_request = Legion::LLM::Inference::Request.build(
+                id:       request_id,
+                messages: messages,
+                routing:  { model: model },
+                tools:    tool_declarations,
+                caller:   effective_caller,
+                stream:   streaming,
+                cache:    { strategy: :default, cacheable: true }
+              )
+
+              executor = Legion::LLM::Inference::Executor.new(inference_request)
+
+              if streaming
+                content_type 'text/event-stream'
+                headers 'Cache-Control'     => 'no-cache',
+                        'Connection'        => 'keep-alive',
+                        'X-Accel-Buffering' => 'no'
+
+                stream do |out|
+                  Responses.stream_response(out, executor, request_id: request_id, model: model)
+                rescue StandardError => e
+                  handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.responses.stream', request_id: request_id)
+                  out << "event: error\ndata: #{Legion::JSON.dump({ type: 'server_error', message: e.message })}\n\n"
+                end
+              else
+                pipeline_response = executor.call
+                response_body = Responses.format_response(pipeline_response, request_id: request_id, model: model)
+
+                log.info("[llm][api][openai][responses] action=complete request_id=#{request_id} model=#{response_body[:model]}")
+                content_type :json
+                status 200
+                Legion::JSON.dump(response_body)
+              end
+            rescue Legion::LLM::AuthError => e
+              handle_exception(e, level: :error, handled: true, operation: 'llm.api.openai.responses.auth')
+              halt 401, { 'Content-Type' => 'application/json' },
+                   Legion::JSON.dump({ error: { message: e.message, type: 'authentication_error' } })
+            rescue Legion::LLM::RateLimitError => e
+              handle_exception(e, level: :warn, handled: true, operation: 'llm.api.openai.responses.rate_limit')
+              halt 429, { 'Content-Type' => 'application/json' },
+                   Legion::JSON.dump({ error: { message: e.message, type: 'rate_limit_error' } })
+            rescue Legion::LLM::ProviderDown, Legion::LLM::ProviderError => e
+              handle_exception(e, level: :error, handled: true, operation: 'llm.api.openai.responses.provider')
+              halt 502, { 'Content-Type' => 'application/json' },
+                   Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
+            rescue StandardError => e
+              handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.responses')
+              halt 500, { 'Content-Type' => 'application/json' },
+                   Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
+            end
+          end
+
+          def self.normalize_input_array(input)
+            input.filter_map do |item|
+              item = item.transform_keys(&:to_sym) if item.respond_to?(:transform_keys)
+
+              case item[:type]&.to_s
+              when 'function_call_output'
+                { role: 'tool', tool_call_id: item[:call_id], content: item[:output].to_s }
+              else
+                role = item[:role]&.to_s
+                next unless role
+
+                content = item[:content]
+                content = content.to_s if content && !content.is_a?(Array)
+                { role: role, content: content }.compact
+              end
+            end
+          end
+
+          def self.build_tool_declarations(tools)
+            return [] if tools.nil? || !tools.is_a?(Array) || tools.empty?
+
+            tools.filter_map do |tool|
+              t = tool.respond_to?(:transform_keys) ? tool.transform_keys(&:to_sym) : tool
+              fn = t[:function] || t
+              fn = fn.transform_keys(&:to_sym) if fn.respond_to?(:transform_keys)
+              next unless fn[:name].to_s.length.positive?
+
+              Legion::LLM::Types::ToolDefinition.build(
+                name:        fn[:name].to_s,
+                description: fn[:description].to_s,
+                parameters:  fn[:parameters] || {},
+                source:      { type: :client, executable: true }
+              )
+            rescue StandardError => e
+              handle_exception(e, level: :warn, handled: true, operation: 'llm.api.openai.responses.build_tool')
+              nil
+            end
+          end
+
+          def self.format_response(pipeline_response, request_id:, model:)
+            routing = pipeline_response.routing || {}
+            tokens = pipeline_response.tokens || {}
+            raw_msg = pipeline_response.message
+            content = raw_msg.is_a?(Hash) ? (raw_msg[:content] || raw_msg['content']).to_s : raw_msg.to_s
+            resolved_model = (routing[:model] || routing['model'] || model).to_s
+
+            output = []
+
+            tool_calls = build_output_tool_calls(pipeline_response)
+            output.concat(tool_calls)
+
+            output << {
+              type:    'message',
+              id:      "msg_#{SecureRandom.hex(12)}",
+              role:    'assistant',
+              content: [{ type: 'output_text', text: content }],
+              status:  'completed'
+            }
+
+            input_tokens = extract_token(tokens, :input)
+            output_tokens = extract_token(tokens, :output)
+
+            {
+              id:         request_id,
+              object:     'response',
+              created_at: Time.now.to_i,
+              model:      resolved_model,
+              output:     output,
+              usage:      {
+                input_tokens:  input_tokens,
+                output_tokens: output_tokens,
+                total_tokens:  input_tokens.to_i + output_tokens.to_i
+              },
+              status:     'completed'
+            }
+          end
+
+          def self.stream_response(out, executor, request_id:, model:)
+            out << "event: response.created\ndata: #{Legion::JSON.dump({ id: request_id, object: 'response', status: 'in_progress' })}\n\n"
+
+            msg_id = "msg_#{SecureRandom.hex(12)}"
+            item_event = { type: 'message', id: msg_id, role: 'assistant', content: [], status: 'in_progress' }
+            out << "event: response.output_item.added\ndata: #{Legion::JSON.dump({ output_index: 0, item: item_event })}\n\n"
+
+            full_text = +''
+
+            pipeline_response = executor.call_stream do |chunk|
+              text = chunk.respond_to?(:content) ? chunk.content.to_s : chunk.to_s
+              next if text.empty?
+
+              full_text << text
+              delta_event = { content_index: 0, delta: text }
+              out << "event: response.output_text.delta\ndata: #{Legion::JSON.dump(delta_event)}\n\n"
+            end
+
+            routing = pipeline_response.routing || {}
+            tokens = pipeline_response.tokens || {}
+            resolved_model = (routing[:model] || routing['model'] || model).to_s
+            input_tokens = extract_token(tokens, :input)
+            output_tokens = extract_token(tokens, :output)
+
+            out << "event: response.output_text.done\ndata: #{Legion::JSON.dump({ content_index: 0, text: full_text })}\n\n"
+            done_item = {
+              output_index: 0,
+              item:         { type: 'message', id: msg_id, role: 'assistant',
+                              content: [{ type: 'output_text', text: full_text }], status: 'completed' }
+            }
+            out << "event: response.output_item.done\ndata: #{Legion::JSON.dump(done_item)}\n\n"
+
+            done_data = {
+              id:     request_id,
+              object: 'response',
+              model:  resolved_model,
+              status: 'completed',
+              usage:  {
+                input_tokens:  input_tokens,
+                output_tokens: output_tokens,
+                total_tokens:  input_tokens.to_i + output_tokens.to_i
+              }
+            }
+            out << "event: response.completed\ndata: #{Legion::JSON.dump(done_data)}\n\n"
+
+            log.info("[llm][api][openai][responses] action=stream_complete request_id=#{request_id} model=#{resolved_model}")
+          end
+
+          def self.build_output_tool_calls(pipeline_response)
+            tools_data = pipeline_response.respond_to?(:tools) ? pipeline_response.tools : nil
+            return [] unless tools_data.is_a?(Array) && !tools_data.empty?
+
+            tools_data.filter_map do |tc|
+              name = tc.respond_to?(:name) ? tc.name : (tc[:name] || tc['name'])
+              args = tc.respond_to?(:arguments) ? tc.arguments : (tc[:arguments] || tc['arguments'] || {})
+              tc_id = tc.respond_to?(:id) ? tc.id : (tc[:id] || tc['id'] || "call_#{SecureRandom.hex(8)}")
+              next unless name
+
+              {
+                type:      'function_call',
+                id:        "fc_#{SecureRandom.hex(12)}",
+                call_id:   tc_id,
+                name:      name.to_s,
+                arguments: args.is_a?(String) ? args : Legion::JSON.dump(args),
+                status:    'completed'
+              }
+            end
+          end
+
+          def self.extract_token(tokens, key)
+            return 0 if tokens.nil?
+            return (tokens[key] || tokens[key.to_s] || 0).to_i if tokens.is_a?(Hash)
+
+            method_name = { input: :input_tokens, output: :output_tokens }[key]
+            return tokens.public_send(method_name).to_i if method_name && tokens.respond_to?(method_name)
+
+            0
+          end
+        end
+      end
+    end
+  end
+end

data/lib/legion/llm/api.rb

@@ -15,6 +15,7 @@ require_relative 'api/translators/openai_response'
 require_relative 'api/openai/chat_completions'
 require_relative 'api/openai/models'
 require_relative 'api/openai/embeddings'
+require_relative 'api/openai/responses'
 require_relative 'api/translators/anthropic_request'
 require_relative 'api/translators/anthropic_response'
 require_relative 'api/anthropic/messages'
@@ -41,6 +42,7 @@ module Legion
         OpenAI::ChatCompletions.registered(app)
         OpenAI::Models.registered(app)
         OpenAI::Embeddings.registered(app)
+        OpenAI::Responses.registered(app)
         Anthropic::Messages.registered(app)
         log.debug('[llm][api] all routes registered')
       end

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.9.32'
+    VERSION = '0.9.33'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.9.32
+  version: 0.9.33
 platform: ruby
 authors:
 - Esity
@@ -204,6 +204,7 @@ files:
 - lib/legion/llm/api/openai/chat_completions.rb
 - lib/legion/llm/api/openai/embeddings.rb
 - lib/legion/llm/api/openai/models.rb
+- lib/legion/llm/api/openai/responses.rb
 - lib/legion/llm/api/translators/anthropic_request.rb
 - lib/legion/llm/api/translators/anthropic_response.rb
 - lib/legion/llm/api/translators/openai_request.rb