legion-llm 0.9.51 → 0.9.53

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e0cae7c608acb8fe3f09852e7833639c08998bb944b2df026d9c1999c03ff017
-  data.tar.gz: bbea922035cf6f38eb43139ea5d33deaca70cbbea8693edccb3811bdc5f43608
+  metadata.gz: 9594126ab3b86d52c27c53288196bafe9951a31953e027db66834e9d3b97baba
+  data.tar.gz: e0b98b99f06be07d64394d55ef86cc906c784178f1b67cb30cac0778c2afcf97
 SHA512:
-  metadata.gz: cc620102bcfdbd73387ba3da2e31e80e4fd9c9b9fd3ceeb85b00417972deeda55bbc427702df3a86ec7a2d3f07be34f99383bf9141b79679e394e66c45eda7c1
-  data.tar.gz: 4f5b8e4739873d147be2ddfed81c6a04297a016f9c7c4c143b6ad0409f61a9f75c39a14c6de4b30f37119b4e5aa577e4faba813917da0a9c771c016500e079a2
+  metadata.gz: 4b467420153c061faffba784a1578b2041d281f94189bc82e0d0555d8cf3fb6132ae22390a46e81598c81ff9d7df0c938ae7b514f54579266636101c158c4334
+  data.tar.gz: b6087830e389876412a55464548cb19b55b15b2a74ecdc9cb6e7f638454c8a8001bca26efce94aed0285f747a368b8fdc19195e0525da2305bdfd1655cfbf30f

data/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Legion LLM Changelog
 
+## [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
+- Discovery: `verify_embedding` now checks `model_available?` for Ollama instead of blindly returning true — prevents `can_embed?` from reporting true when the embedding model (e.g. `mxbai-embed-large`) hasn't been pulled on the local node
+- Discovery: `detect_embedding_from_registry` now calls `verify_embedding` before setting `@can_embed = true`, closing a gap where registry-declared capability metadata was trusted without verifying the model exists locally
+
 ## [0.9.51] - 2026-05-23
 
 ### Changed

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
 
@@ -121,8 +130,6 @@ 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/discovery.rb

@@ -266,6 +266,12 @@ module Legion
             return false
           end
 
+          unless verify_embedding(provider, resolved)
+            log.debug '[llm][discovery] action=detect_embedding_from_registry verify_failed ' \
+                      "provider=#{provider} model=#{resolved} — falling through to legacy probe"
+            return false
+          end
+
           @embedding_provider = provider
           @embedding_model    = resolved
           @embedding_instance = instance
@@ -324,16 +330,9 @@ module Legion
 
         def verify_embedding(provider, model)
           log.debug "[llm][discovery] verify_embedding provider=#{provider} model=#{model}"
-          return true if provider == :ollama
-          return true if provider == :azure
-          return false unless provider_supports_embeddings?(provider)
           return true unless model
 
-          start_time = Time.now
-          Call::Dispatch.call(provider: provider, capability: :embed, model: model, text: 'health check')
-          elapsed = ((Time.now - start_time) * 1000).round
-          log.info "[llm][discovery] embedding health check ok provider=#{provider} model=#{model} elapsed_ms=#{elapsed}"
-          true
+          model_available?(model, provider: provider)
         rescue StandardError => e
           handle_exception(e, level: :warn, operation: 'llm.discovery.verify_embedding', provider: provider, model: model)
           false

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.9.51'
+    VERSION = '0.9.53'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.9.51
+  version: 0.9.53
 platform: ruby
 authors:
 - Esity