@fugood/bricks-project 2.22.0-beta.7 → 2.22.0-beta.9

package/compile/action-name-map.ts

@@ -743,6 +743,10 @@ export const templateActionNameMap = {
       mcpVariables: 'GENERATOR_ASSISTANT_MCP_VARIABLES',
       role: 'GENERATOR_ASSISTANT_ROLE',
     },
+    GENERATOR_ASSISTANT_SUMMARY_MESSAGES: {
+      summaryMessages: 'GENERATOR_ASSISTANT_SUMMARY_MESSAGES',
+      summarySessionKey: 'GENERATOR_ASSISTANT_SUMMARY_SESSION_KEY',
+    },
   },
   GENERATOR_VECTOR_STORE: {
     GENERATOR_VECTOR_STORE_RESET: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/bricks-project",
-  "version": "2.22.0-beta.7",
+  "version": "2.22.0-beta.9",
   "main": "index.ts",
   "scripts": {
     "build": "node scripts/build.js"
@@ -14,5 +14,5 @@
     "lodash": "^4.17.4",
     "uuid": "^8.3.1"
   },
-  "gitHead": "b205bf371305e9d4929f05c0c310ff2362a43739"
+  "gitHead": "286abe17fdd111c728fd74e3d98822d7e11d718e"
 }

package/types/generators.ts

@@ -4497,6 +4497,8 @@ Default property:
     isListening?: () => Data
     /* Last error of HTTP server */
     lastError?: () => Data
+    /* MCP server endpoint URL */
+    endpoint?: () => Data
     /* Connected remotes (Session ID) */
     connectedRemotes?: () => Data
     /* Last resource request ({ name: string, uri: string, params: object }) */
@@ -4525,6 +4527,7 @@ export type GeneratorMCPServer = Generator &
                   outlet:
                     | 'isListening'
                     | 'lastError'
+                    | 'endpoint'
                     | 'connectedRemotes'
                     | 'lastResourceRequest'
                     | 'lastToolCall'
@@ -6174,6 +6177,7 @@ Default property:
   "useMmap": true,
   "cacheKType": "f16",
   "cacheVType": "f16",
+  "ctxShift": true,
   "transformScriptEnabled": false,
   "transformScriptCode": "\/\* Global variable: inputs = { prompt, messages, variables } \*\/\nreturn inputs.prompt",
   "transformScriptVariables": {},
@@ -6259,6 +6263,8 @@ Default property:
     cacheKType?: 'f16' | 'f32' | 'q8_0' | 'q4_0' | 'q4_1' | 'iq4_nl' | 'q5_0' | 'q5_1' | DataLink
     /* KV cache data type for the V (Default: f16) */
     cacheVType?: 'f16' | 'f32' | 'q8_0' | 'q4_0' | 'q4_1' | 'iq4_nl' | 'q5_0' | 'q5_1' | DataLink
+    /* Enable context shift */
+    ctxShift?: boolean | DataLink
     /* Enable Transform Script for processing the prompt */
     transformScriptEnabled?: boolean | DataLink
     /* Code of Transform Script */
@@ -6277,8 +6283,10 @@ Default property:
     sessionRemain?: number | DataLink
     /* TODO:loran_gqarms_norm_epsrope_freq_baserope_freq_scale */
     completionMode?: 'auto' | 'chat' | 'text' | DataLink
-    /* Tools for chat mode */
-    completionTools?: {} | DataLink
+    /* Tools for chat mode using OpenAI-compatible function calling format  
+ Format: Array of objects with {type, function: {name, description, parameters}} structure  
+ See: https://platform.openai.com/docs/guides/function-calling */
+    completionTools?: Array<{} | DataLink> | DataLink
     /* Enable parallel tool calls */
     completionParallelToolCalls?: boolean | DataLink
     /* Tool choice for chat mode */
@@ -6534,8 +6542,10 @@ Default property:
     stopWords?: Array<string | DataLink> | DataLink
     /* Tool call parser */
     toolCallParser?: 'llama3_json' | 'mistral' | DataLink
-    /* Tools for chat mode */
-    tools?: {} | DataLink
+    /* Tools for chat mode using OpenAI-compatible function calling format  
+ Format: Array of objects with {type, function: {name, description, parameters}} structure  
+ See: https://platform.openai.com/docs/guides/function-calling */
+    tools?: Array<{} | DataLink> | DataLink
     /* Tool choice for chat mode */
     toolChoice?: 'none' | 'auto' | 'required' | DataLink
     /* Enable parallel tool calls */
@@ -6678,7 +6688,10 @@ Default property:
   "apiEndpoint": "https://api.openai.com/v1",
   "model": "gpt-4o-mini",
   "completionMessages": [
-    null
+    {
+      "role": "system",
+      "content": "You are a helpful assistant."
+    }
   ],
   "completionMaxTokens": 1024,
   "completionTemperature": 1,
@@ -6714,8 +6727,10 @@ Default property:
             }
         >
       | DataLink
-    /* Tools for chat mode */
-    completionTools?: {} | DataLink
+    /* Tools for chat mode following OpenAI function calling format  
+ Format: Array of objects with {type, function: {name, description, parameters}} structure  
+ See: https://platform.openai.com/docs/guides/function-calling */
+    completionTools?: Array<{} | DataLink> | DataLink
     /* Enable parallel tool calls */
     completionParallelToolCalls?: boolean | DataLink
     /* Tool choice for chat mode */
@@ -7224,16 +7239,50 @@ export type GeneratorAssistantActionInsertMcpResource = ActionWithParams & {
   >
 }
 
+/* Summarize messages based on the conversation  
+
+ Note: Summary uses the same LLM context size, so it is recommended only to use it when the system prompt (in Initial Messages) is long,  otherwise it may still fail when the context is full (Ctx Shift is NO). */
+export type GeneratorAssistantActionSummaryMessages = ActionWithParams & {
+  __actionName: 'GENERATOR_ASSISTANT_SUMMARY_MESSAGES'
+  params?: Array<
+    | {
+        input: 'summaryMessages'
+        value?: Array<any> | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'summarySessionKey'
+        value?: string | DataLink | EventProperty
+        mapping?: string
+      }
+  >
+}
+
 interface GeneratorAssistantDef {
   /*
 Default property:
 {
   "initialMessages": [
-    null
+    {
+      "role": "system",
+      "content": "You are a helpful assistant."
+    }
   ],
   "cacheMessages": false,
   "llmLivePolicy": "only-in-use",
   "llmSessionKey": "default-assistant",
+  "llmAutoSummaryMessages": false,
+  "llmSummaryMessages": [
+    {
+      "role": "system",
+      "content": "You are a helpful assistant specialized in summarizing conversations. Create a concise summary of the conversation that captures the key points while maintaining important context. The summary should be clear, accurate, and briefer than the original conversation."
+    },
+    {
+      "role": "user",
+      "content": "Please summarize the following conversation into a concise system message that can replace the previous conversation context while maintaining all important information. Here is the conversation to summarize:\n\n"
+    }
+  ],
+  "llmSummarySessionKey": "assistant-default-summary",
   "fileSearchEnabled": false,
   "fileSearchLivePolicy": "only-in-use",
   "sttEnabled": true,
@@ -7257,12 +7306,28 @@ Default property:
       | DataLink
     /* Whether to cache messages */
     cacheMessages?: boolean | DataLink
-    /* LLM Generator (Currently only support `LLM (GGML)` generator) */
+    /* LLM Generator (Supports `LLM (GGML)` and `OpenAI LLM` generators) */
     llmGeneratorId?: string | DataLink
     /* LLM Live Policy. If the policy is `only-in-use`, the LLM context will be released when the assistant is not in use. */
     llmLivePolicy?: 'only-in-use' | 'manual' | DataLink
     /* LLM main session key */
     llmSessionKey?: string | DataLink
+    /* Auto Summary Messages (Automatically summarize messages when the LLM context is full or content gets truncated, currently only supported with LLM (GGML) generators)  
+
+ Note: Summary uses the same LLM context size, so it is recommended only to use it when the system prompt (in Initial Messages) is long,  otherwise it may still fail when the context is full (Ctx Shift is NO). */
+    llmAutoSummaryMessages?: boolean | DataLink
+    /* Summary Messages (Messages used for summarization prompt, conversation will be appended to the last message) */
+    llmSummaryMessages?:
+      | Array<
+          | DataLink
+          | {
+              role?: string | DataLink
+              content?: string | DataLink
+            }
+        >
+      | DataLink
+    /* Summary Session Key (Custom session key for summarization) */
+    llmSummarySessionKey?: string | DataLink
     /* File Search (Vector Store) Enabled */
     fileSearchEnabled?: boolean | DataLink
     /* File Search (Vector Store) Generator */
@@ -7275,13 +7340,13 @@ Default property:
     fileSearchThreshold?: number | DataLink
     /* File Search Ignore Threshold. (Default: false) */
     fileSearchIgnoreThreshold?: boolean | DataLink
-    /* STT Generator use for transcribing audio message (Currently only support `STT (GGML)` generator) */
+    /* STT Generator use for transcribing audio message (Supports `STT (GGML)` generators) */
     sttGeneratorId?: string | DataLink
     /* STT Enabled */
     sttEnabled?: boolean | DataLink
     /* STT Live Policy. If the policy is `only-in-use`, the STT context will be released when the assistant is not in use. */
     sttLivePolicy?: 'only-in-use' | 'manual' | DataLink
-    /* TTS Generator use for generating LLM response audio message (Currently only support `TTS (ONNX)` generator) */
+    /* TTS Generator use for generating LLM response audio message (Supports `TTS (ONNX)` and `OpenAI TTS` generators) */
     ttsGeneratorId?: string | DataLink
     /* TTS Enabled */
     ttsEnabled?: boolean | DataLink

package/utils/event-props.ts

@@ -770,6 +770,8 @@ export const templateEventPropsMap = {
       'GENERATOR_LLM_COMPLETION_RESULT', // type: string
       'GENERATOR_LLM_COMPLETION_TOOL_CALLS', // type: array
       'GENERATOR_LLM_COMPLETION_FULL_CONTEXT', // type: string
+      'GENERATOR_LLM_COMPLETION_IS_CONTEXT_FULL', // type: bool
+      'GENERATOR_LLM_COMPLETION_IS_TRUNCATED', // type: bool
       'GENERATOR_LLM_COMPLETION_RESULT_DETAILS', // type: object
     ],
     onCompletionFunctionCall: [