chore(*): culmulative AI GW fixes 20251028

changelog/unreleased/kong/feat-add-llm-bedrock-agent-runtime-sdk.yml

@@ -0,0 +1,3 @@
+message: "**llm**: Added support for Bedrock agent SDK, including rerand, converse, converse-stream, retrieveAndGenerate, and retrieveAndGenerate-stream."
+type: "feature"
+scope: "Plugin"

changelog/unreleased/kong/feat-add-support-for-huggingface-serverless-inference-provider.yml

@@ -0,0 +1,4 @@
+message: |
+  **ai-proxy**: Added support for HuggingFace's new serverless API in the AI Proxy plugin, enabling seamless integration and improved compatibility.
+type: "feature"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-analytics.yml

@@ -0,0 +1,3 @@
+message: Fixed an issue where some of ai metrics was missed in analytics
+scope: Plugin
+type: bugfix

changelog/unreleased/kong/fix-ai-bedrock-tool-use.yml

@@ -0,0 +1,4 @@
+message: >
+  **AI Plugins**: Fixed an issue where Bedrock (Converse API) didn't properly parse multiple `toolUse` tool calls in one turn, on some models.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-ai-driver-upstream-scheme-400.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-driver**: Fixed an issue where AI Proxy and AI Proxy Advanced generated incorrect default ports for upstream schemes, leading to 400 errors.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-ai-gcp-model-armor-floor-blocked.yml

@@ -0,0 +1,4 @@
+message: >
+  **AI Plugins**: Fixed an issue where the Gemini provider would not correctly return Model Armor 'Floor' blocking responses to the caller.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-ai-log-for-huge-request-payload.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where large request payload was not logged.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-plugins-using-capture-groups.yml

@@ -0,0 +1,4 @@
+message: |
+  **ai-proxy**: Fixed an issue where patterns using multiple capture groups (e.g., `$(group1)/$(group2)`) failed to extract expected matches.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-anthropic-completions-stop-turn-mapping.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where the Anthropic provider mapped the wrong stop_reason to Chat and Completions responses.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-anthropic-streaming-tool-use.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where the Anthropic provider failed to stream function call responses.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-ai-proxy-anthropic-streaming-truncation.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where the Anthropic provider may truncate tokens in streaming responses.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-ai-proxy-anthropic-structured-output.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue Structured Output did not work correctly with Anthropic Claude models."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-bedrock-streaming-structured-output.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue where structured output did not work for the Bedrock provider when SSE streaming responses."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-bedrock-structured-output.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue Structured Output did not work correctly with Bedrock (Converse API) models."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-gemini-anthropic-rawpredict.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue where the Gemini provider could not use Anthropic 'rawPredict' endpoint models hosted in Vertex."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-gemini-id-created-missing.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue where `id`, `created`, and/or `finish_reason` fields were missing from or incorrect in Gemini responses."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-gemini-model-missing.yml

@@ -0,0 +1,3 @@
+message: "**ai-proxy**: Fixed an issue where `model` field was missing in OpenAI format response from Gemini provider in some situations."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-gemini-tool-calls.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where the Gemini provider did not properly handle multiple tool calls across different turns.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-latency-for-streaming.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where latency metric was not implemented for streaming responses.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-structured-output-empty-array.yml

@@ -0,0 +1,3 @@
+message: "**ai-proxy**: Fixed an issue when empty array in structured output was encoded as empty object."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-proxy-structured-output.yml

@@ -0,0 +1,3 @@
+message: "**AI Plugins**: Fixed an issue where Structured-Output requests were ignored."
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-ai-retry-plugin-ctx-incorrect-ns.yml

@@ -0,0 +1,4 @@
+message: |
+  **ai-proxy-advanced**: Fixed an issue where ai-retry phase was not correctly setting the namespace in the kong.plugin.ctx, causing ai-proxy-advanced balancer retry first target more than once.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-aws-stream-parse-incomplete-frame.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where aws stream parser didn't parse correctly when the frame was incomplete.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-gemini-vertex.yml

@@ -0,0 +1,4 @@
+message: |
+  **Semantic Plugins*: Fixed an issue where Gemini Vertex AI embeddings failed due to incorrect URL construction and response parsing.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-huggingface-embedding-response-inproperly-parsed.yml

@@ -0,0 +1,4 @@
+message: |
+  **ai-proxy**: Fixed an issue where HuggingFace embedding driver were incorrectly parsed responses from embedding API
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-json-array-iterator.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where JSON array data is not correctly parsed.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-llm-failover-bedrock.yml

@@ -0,0 +1,4 @@
+message: >
+  Fixed an issue where AI Proxy and AI Proxy Advanced can't properly failover to a Bedrock target.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-llm-sse-parser-with-truncated-sse.yml

@@ -0,0 +1,4 @@
+message: >
+  Fixed an issue where AI Proxy and AI Proxy Advanced might produce duplicate content in the response when the SSE event was truncated.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-llm-sse-parser-with-truncated-sse2.yml

@@ -0,0 +1,4 @@
+message: >
+  Fixed an issue where AI Proxy and AI Proxy Advanced might drop content in the response when the SSE event was truncated.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix-sse-terminator.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where SSE terminator may not have correct ending characters.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix-wrong-body-for-observability.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where the response body for observability may be larger than the real one because of the stale data.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix_ai-proxy-huggingface-request-extra-parameters-not-allowd.yml

@@ -0,0 +1,4 @@
+message: |
+  **AI Proxy**: Fixed an issue where extra inputs were not permitted for huggingface inference provider
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/fix_ai_embeddings_titan_embed_dimensions.yml

@@ -0,0 +1,4 @@
+message: >
+  **AI Plugins**: Fixed unable to use Titan Text Embeddings models.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix_ai_gemini_model_garden.yml

@@ -0,0 +1,4 @@
+message: >
+  "AI Proxy Advanced: Fixed an issue where gemini provider not support model garden"
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix_ai_prompt_decorator_content_length.yml

@@ -0,0 +1,4 @@
+message: |
+  Fixed an issue where the `ai-prompt-decorator` plugin prompt field was too short.
+type: feature
+scope: Plugin

changelog/unreleased/kong/fix_ai_prompt_decorator_missing_fields.yml

@@ -0,0 +1,4 @@
+message: |
+  **ai-prompt-decorator**: Fixed an issue where the plugin would miss the model, temperature, and other user-defined fields.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix_ai_proxy_gemini_driver_return_403.yml

@@ -0,0 +1,4 @@
+message: >
+  **ai-proxy**: Fixed an issue where AI Proxy returns 403 using gemini provider.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix_ai_proxy_ollama_rag_and_tools.yml

@@ -0,0 +1,4 @@
+message: >
+  **ollama**: Fixed an issue where 'tools' and 'RAG' decoration did not work for Ollama (llama2) provider.
+type: bugfix
+scope: Plugin

changelog/unreleased/kong/fix_mistral_remove_seed.yml

@@ -0,0 +1,4 @@
+message: |
+  **AI Proxy Advanced**: Fixed an issue where Mistral models would return `Unsupported field: seed` when using some inference libraries.
+type: "bugfix"
+scope: "Plugin"

changelog/unreleased/kong/perf-parse-sse-chunk.yml

@@ -0,0 +1,3 @@
+message: "**ai-proxy**: Implemented a faster SSE parser."
+type: "performance"
+scope: "Plugin"

kong-latest.rockspec

@@ -664,6 +664,7 @@ build = {
     ["kong.llm.plugin.shared-filters.parse-json-response"] = "kong/llm/plugin/shared-filters/parse-json-response.lua",
     ["kong.llm.plugin.shared-filters.parse-request"] = "kong/llm/plugin/shared-filters/parse-request.lua",
     ["kong.llm.plugin.shared-filters.parse-sse-chunk"] = "kong/llm/plugin/shared-filters/parse-sse-chunk.lua",
+    ["kong.llm.plugin.shared-filters.save-request-body"] = "kong/llm/plugin/shared-filters/save-request-body.lua",
     ["kong.llm.plugin.shared-filters.serialize-analytics"] = "kong/llm/plugin/shared-filters/serialize-analytics.lua",
 
     ["kong.llm.adapters.bedrock"] = "kong/llm/adapters/bedrock.lua",

kong/llm/adapters/bedrock.lua

@@ -1,6 +1,7 @@
 local cjson = require("cjson.safe")
 local fmt = string.format
 
+
 local _BedrockAdapter = {}
 
 _BedrockAdapter.role_map = {
@@ -305,10 +306,51 @@ function _BedrockAdapter:to_kong_req(bedrock_table, kong)
     openai_table.tools = self:extract_tools(bedrock_table.toolConfig.tools)
   end
 
+  local url = kong.request.get_path()
+  if url:find("/rerank", 1, true) then      
+    self.forward_path = "/rerank"
+    openai_table.messages = bedrock_table.queries
+    return openai_table
+  elseif url:find("/retrieveAndGenerateStream", 1, true) then
+      self.forward_path = "/retrieveAndGenerateStream"
+      openai_table.prompt = bedrock_table.input and bedrock_table.input.text
+      openai_table.stream = true
+      return openai_table
+  elseif url:find("/retrieveAndGenerate", 1, true) then
+    self.forward_path = "/retrieveAndGenerate"
+    openai_table.prompt = bedrock_table.input and bedrock_table.input.text
+    return openai_table
+  elseif url:find('converse-stream',1, true) then
+    self.forward_path = "/model/%s/converse-stream"
+    openai_table.stream = true
+    return openai_table
+  elseif url:find("converse",1, true) then
+    self.forward_path = "/model/%s/converse"
+    return openai_table
+  end
   return openai_table
 end
 
 
+function _BedrockAdapter:get_forwarded_path(model)
+  if not self.forward_path then
+    return
+  end
+
+  -- if the forward path is a string, it means we have to format it with the model name
+  if type(self.forward_path) == "string" then
+    local forward_path = fmt(self.forward_path, model)
+    if self.forward_path == "/rerank" or 
+        self.forward_path == "/retrieveAndGenerate" or
+        self.forward_path == "/retrieveAndGenerateStream" then
+          return forward_path, "bedrock_agent"
+    end
+    return forward_path, "bedrock"
+  end
+
+  return
+end
+
 -- for unit tests
 if _G.TEST then
   _BedrockAdapter._set_kong = function(this_kong)