legion-llm 0.9.52 → 0.9.54

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: '08d2a4d13f38bdf0c305da0965e688c317deffdad792796daf66f0168bddd37a'
-  data.tar.gz: 9c36b44103590f45b9947fee8c3dfaf9651fcec57c193b7c7b4dbfe839e57aed
+  metadata.gz: '09bf7eb9fe4c93ccba0bb574864c66a518225327399b47a30dd08aa148ac3b74'
+  data.tar.gz: a54854addf081e387d94ed9fd3ba67f826df9b64ec3b042b2767a7e5a441d715
 SHA512:
-  metadata.gz: 5fec699d874dff0e91c83735becbfb825657c1d9da13e7034c8e6a843dba481e85a30c7975c959282ca8da5e9140eeeff92489a3e32fdad164685d33f83f9e42
-  data.tar.gz: 3e149f5061993af9a16f4644b83385a0c3a60fcaa394a2454c88cdecd99890d4fd083c08491fe22715739507875bb544b61747947d2805cc198ed72f9cabde3c
+  metadata.gz: 69fa173952297d7da6410c101c9b9e13548514db85075188fdef153d0f9340f50c314e3f633432e5481c5a4d979468ea84dc3b24edff84d4d728a6d7df7f94c0
+  data.tar.gz: 52852f542515ec121bf1d9851506c3866e70cae5a551f034777f15d9c68fc7886888eb6d9c5cb9a9f58e18016d84d92d4dd010a58cc7c312b2c683bfb477815d

data/CHANGELOG.md

@@ -1,5 +1,16 @@
 # Legion LLM Changelog
 
+## [0.9.54] - 2026-05-29
+
+### Fixed
+- API: OpenAI-compatible streaming responses now include `usage` (`prompt_tokens`, `completion_tokens`, `total_tokens`) in the final done-chunk, allowing the Vercel AI SDK `step-finish` event to propagate token counts to clients (fixes empty Tokens display in Kai's response info popup)
+
+## [0.9.53] - 2026-05-29
+
+### Added
+- API: OpenAI-compatible endpoints (`/v1/chat/completions`, `/v1/models`, `/v1/embeddings`, `/v1/responses`) are now also available under the `/api/llm/inference/v1/` prefix, allowing Mastra `openai-compatible` providers to use `http://127.0.0.1:4567/api/llm/inference` as the base URL — consistent with the Claude and Codex client routing patches in legion-interlink
+- API: auth `before` filter extended to cover `/api/llm/inference/v1/*` in addition to `/v1/*`
+
 ## [0.9.52] - 2026-05-27
 
 ### Fixed

data/lib/legion/llm/api/auth.rb

@@ -9,9 +9,9 @@ module Legion
         extend Legion::Logging::Helper
 
         def self.registered(app)
-          log.debug('[llm][api][auth] registering /v1/* before filter')
+          log.debug('[llm][api][auth] registering /v1/* and /api/llm/inference/v1/* before filters')
 
-          app.before '/v1/*' do
+          auth_check = proc do
             log.debug("[llm][api][auth] before filter action=check path=#{request.path_info}")
             next unless auth_enabled?
 
@@ -27,6 +27,10 @@ module Legion
             log.debug("[llm][api][auth] action=authorized path=#{request.path_info}")
           end
 
+          app.before('/api/llm/inference/v1/*', &auth_check)
+
+          app.before('/v1/*', &auth_check)
+
           app.helpers do
             define_method(:auth_enabled?) do
               Legion::LLM::Settings.value(:api, :auth, :enabled) == true
@@ -53,7 +57,7 @@ module Legion
             end
           end
 
-          log.debug('[llm][api][auth] /v1/* before filter registered')
+          log.debug('[llm][api][auth] /v1/* and /api/llm/inference/v1/* before filters registered')
         rescue StandardError => e
           handle_exception(e, level: :error, handled: false, operation: 'llm.api.auth.register')
         end

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

@@ -11,10 +11,19 @@ module Legion
         module ChatCompletions
           extend Legion::Logging::Helper
 
-          def self.registered(app) # rubocop:disable Metrics/AbcSize,Metrics/MethodLength
-            log.debug('[llm][api][openai][chat_completions] registering POST /v1/chat/completions')
+          def self.registered(app)
+            log.debug('[llm][api][openai][chat_completions] registering POST /v1/chat/completions + /api/llm/inference/v1/chat/completions')
 
-            app.post '/v1/chat/completions' do # rubocop:disable Metrics/BlockLength
+            handler = build_handler
+
+            app.post('/v1/chat/completions') { instance_exec(&handler) }
+            app.post('/api/llm/inference/v1/chat/completions') { instance_exec(&handler) }
+
+            log.debug('[llm][api][openai][chat_completions] routes registered')
+          end
+
+          def self.build_handler # rubocop:disable Metrics/MethodLength,Metrics/AbcSize
+            proc do # rubocop:disable Metrics/BlockLength
               require_llm!
               body = parse_request_body
 
@@ -82,7 +91,13 @@ module Legion
                     nil,
                     model:         final_model,
                     request_id:    request_id,
-                    finish_reason: tool_calls.empty? ? 'stop' : 'tool_calls'
+                    finish_reason: tool_calls.empty? ? 'stop' : 'tool_calls',
+                    usage:         {
+                      prompt_tokens:     Legion::LLM::API::Translators::OpenAIResponse.extract_token_count(pipeline_response.tokens, :input),
+                      completion_tokens: Legion::LLM::API::Translators::OpenAIResponse.extract_token_count(pipeline_response.tokens, :output),
+                      total_tokens:      Legion::LLM::API::Translators::OpenAIResponse.extract_token_count(pipeline_response.tokens, :input).to_i +
+                                         Legion::LLM::API::Translators::OpenAIResponse.extract_token_count(pipeline_response.tokens, :output).to_i
+                    }
                   )
                   out << "data: #{Legion::JSON.dump(done_chunk)}\n\n"
                   out << "data: [DONE]\n\n"
@@ -121,8 +136,6 @@ module Legion
               halt 500, { 'Content-Type' => 'application/json' },
                    Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
             end
-
-            log.debug('[llm][api][openai][chat_completions] POST /v1/chat/completions registered')
           end
 
           def self.build_openai_tool_classes(tools)

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

@@ -10,9 +10,18 @@ module Legion
           extend Legion::Logging::Helper
 
           def self.registered(app)
-            log.debug('[llm][api][openai][embeddings] registering POST /v1/embeddings')
+            log.debug('[llm][api][openai][embeddings] registering POST /v1/embeddings + /api/llm/inference/v1/embeddings')
 
-            app.post '/v1/embeddings' do
+            handler = build_handler
+
+            app.post('/v1/embeddings') { instance_exec(&handler) }
+            app.post('/api/llm/inference/v1/embeddings') { instance_exec(&handler) }
+
+            log.debug('[llm][api][openai][embeddings] routes registered')
+          end
+
+          def self.build_handler
+            proc do
               require_llm!
               body = parse_request_body
 
@@ -57,8 +66,6 @@ module Legion
               halt 500, { 'Content-Type' => 'application/json' },
                    Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
             end
-
-            log.debug('[llm][api][openai][embeddings] POST /v1/embeddings registered')
           end
         end
       end

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

@@ -11,9 +11,9 @@ module Legion
           extend Legion::Logging::Helper
 
           def self.registered(app)
-            log.debug('[llm][api][openai][models] registering GET /v1/models and GET /v1/models/:id')
+            log.debug('[llm][api][openai][models] registering GET /v1/models + /api/llm/inference/v1/models routes')
 
-            app.get '/v1/models' do
+            list_handler = proc do
               log.debug('[llm][api][openai][models] action=list')
               require_llm!
 
@@ -28,7 +28,7 @@ module Legion
                    Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
             end
 
-            app.get '/v1/models/:id' do
+            get_handler = proc do
               model_id = params[:id]
               log.debug("[llm][api][openai][models] action=get id=#{model_id}")
               require_llm!
@@ -52,7 +52,12 @@ module Legion
                    Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
             end
 
-            log.debug('[llm][api][openai][models] GET /v1/models routes registered')
+            app.get('/v1/models') { instance_exec(&list_handler) }
+            app.get('/api/llm/inference/v1/models') { instance_exec(&list_handler) }
+            app.get('/v1/models/:id') { instance_exec(&get_handler) }
+            app.get('/api/llm/inference/v1/models/:id') { instance_exec(&get_handler) }
+
+            log.debug('[llm][api][openai][models] routes registered')
           end
 
           def self.build_model_list

data/lib/legion/llm/api/translators/openai_response.rb

@@ -57,17 +57,19 @@ module Legion
             }
           end
 
-          def format_stream_chunk(delta_text, model:, request_id:, finish_reason: nil)
+          def format_stream_chunk(delta_text, model:, request_id:, finish_reason: nil, usage: nil)
             choice = { index: 0, delta: {}, finish_reason: finish_reason }
             choice[:delta][:content] = delta_text if delta_text && !delta_text.empty?
 
-            {
+            chunk = {
               id:      "chatcmpl-#{request_id.delete('-')}",
               object:  'chat.completion.chunk',
               created: Time.now.to_i,
               model:   model.to_s,
               choices: [choice]
             }
+            chunk[:usage] = usage if usage
+            chunk
           end
 
           def format_stream_tool_call_chunk(tool_call, model:, request_id:, index:)

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.9.52'
+    VERSION = '0.9.54'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.9.52
+  version: 0.9.54
 platform: ruby
 authors:
 - Esity