zrb 1.8.6__py3-none-any.whl → 1.8.8__py3-none-any.whl

zrb/builtin/llm/llm_ask.py

@@ -141,6 +141,6 @@ if CFG.LLM_ALLOW_ACCESS_SHELL:
 
 if CFG.LLM_ALLOW_ACCESS_INTERNET:
     llm_ask.append_tool(open_web_page, search_wikipedia, search_arxiv)
-    if CFG.SERP_API_KEY != "":
-        llm_ask.append_tool(create_search_internet_tool(CFG.SERP_API_KEY))
+    if CFG.SERPAPI_KEY != "":
+        llm_ask.append_tool(create_search_internet_tool(CFG.SERPAPI_KEY))
     llm_ask.append_tool(get_current_location, get_current_weather)

zrb/builtin/llm/tool/code.py

@@ -4,7 +4,7 @@ from zrb.builtin.llm.tool.file import DEFAULT_EXCLUDED_PATTERNS, is_excluded
 from zrb.builtin.llm.tool.sub_agent import create_sub_agent_tool
 from zrb.context.any_context import AnyContext
 
-_EXTRACT_INFO_SYSTEM_PROMPT = """
+_EXTRACT_INFO_FROM_REPO_SYSTEM_PROMPT = """
 You are an extraction info agent.
 Your goal is to help to extract relevant information to help the main LLM Agent.
 You write your output is in markdown format containing path and relevant information.
@@ -169,7 +169,7 @@ async def _extract_info(
     extract = create_sub_agent_tool(
         tool_name="extract",
         tool_description="extract",
-        system_prompt=_EXTRACT_INFO_SYSTEM_PROMPT,
+        system_prompt=_EXTRACT_INFO_FROM_REPO_SYSTEM_PROMPT,
     )
     extracted_infos = []
     content_buffer = ""

zrb/builtin/llm/tool/file.py

@@ -8,6 +8,28 @@ from zrb.builtin.llm.tool.sub_agent import create_sub_agent_tool
 from zrb.context.any_context import AnyContext
 from zrb.util.file import read_file, read_file_with_line_numbers, write_file
 
+_EXTRACT_INFO_SYSTEM_PROMPT = """
+You are an extraction info agent.
+Your goal is to help to extract relevant information to help the main LLM Agent.
+You write your output is in markdown format containing path and relevant information.
+Extract only information that relevant to main LLM Agent's goal.
+
+Extracted Information format (Use this as reference, extract relevant information only):
+# imports
+- <imported-package>
+- ...
+# variables
+- <variable-type> <variable-name>: <the-purpose-of-the-variable>
+- ...
+# functions
+- <function-name>:
+  - parameters: <parameters>
+  - logic/description: <what-the-function-do-and-how-it-works>
+...
+...
+""".strip()
+
+
 DEFAULT_EXCLUDED_PATTERNS = [
     # Common Python artifacts
     "__pycache__",
@@ -464,13 +486,7 @@ async def analyze_file(ctx: AnyContext, path: str, query: str) -> str:
     _analyze_file = create_sub_agent_tool(
         tool_name="analyze_file",
         tool_description="analyze file with LLM capability",
-        system_prompt="\n".join(
-            [
-                "You are a file analyzer assistant",
-                "Your goal is to help the main asisstant by reading file",
-                "and perform necessary indepth analysis required by the main assistant",
-            ]
-        ),
+        system_prompt=_EXTRACT_INFO_SYSTEM_PROMPT,
         tools=[read_from_file, search_files],
     )
     return await _analyze_file(

zrb/config.py

@@ -245,17 +245,17 @@ class Config:
     @property
     def LLM_MAX_TOKENS_PER_MINUTE(self) -> int:
         """Maximum number of LLM tokens allowed per minute."""
-        return int(os.getenv("LLM_MAX_TOKENS_PER_MINUTE", "120000"))
+        return int(os.getenv("ZRB_LLM_MAX_TOKENS_PER_MINUTE", "120000"))
 
     @property
     def LLM_MAX_TOKENS_PER_REQUEST(self) -> int:
         """Maximum number of tokens allowed per individual LLM request."""
-        return int(os.getenv("LLM_MAX_TOKENS_PER_REQUEST", "4096"))
+        return int(os.getenv("ZRB_LLM_MAX_TOKENS_PER_REQUEST", "30000"))
 
     @property
     def LLM_THROTTLE_SLEEP(self) -> float:
         """Number of seconds to sleep when throttling is required."""
-        return float(os.getenv("LLM_THROTTLE_SLEEP", "1.0"))
+        return float(os.getenv("ZRB_LLM_THROTTLE_SLEEP", "1.0"))
 
     @property
     def LLM_CONTEXT_ENRICHMENT_PROMPT(self) -> str | None:
@@ -321,8 +321,8 @@ class Config:
         return int(os.getenv("ZRB_RAG_MAX_RESULT_COUNT", "5"))
 
     @property
-    def SERP_API_KEY(self) -> str:
-        return os.getenv("SERP_API_KEY", "")
+    def SERPAPI_KEY(self) -> str:
+        return os.getenv("SERPAPI_KEY", "")
 
     @property
     def BANNER(self) -> str:

zrb/llm_config.py

@@ -12,42 +12,55 @@ else:
 from zrb.config import CFG
 
 DEFAULT_SYSTEM_PROMPT = """
-You have access to tools.
-Your goal is to complete the user's task efficiently.
+You have access to tools and two forms of memory: a narrative summary of the long-term conversation and a structured JSON object with key facts.
+Your goal is to complete the user's task efficiently by synthesizing information from both memory types and the current turn.
 Analyze the request and use the available tools proactively to achieve the goal.
 Infer parameters and actions from the context.
-Do not ask for confirmation unless strictly necessary due to ambiguity or
-missing critical information.
+Do not ask for confirmation unless strictly necessary for irreversible actions or due to critical ambiguity.
 Apply relevant domain knowledge and best practices.
 Respond directly and concisely upon task completion or when clarification is essential.
 Make sure to always include all necessary information in your final answer.
-Remember that you have limited set of context.
+Remember that your narrative summary may be condensed; rely on the structured JSON for precise facts when available.
 """.strip()
 
 DEFAULT_PERSONA = """
-You are an expert in various fields including technology, science, history, and more.
+You are a helpful, clear, and precise expert in various fields including technology, science, history, and more.
+As an expert, your goal is to provide accurate information efficiently, getting straight to the point while remaining helpful.
+While you are an expert, if you are uncertain about a fact, state what you know and what you are unsure about.
 """.strip()
 
 # Concise summarization focused on preserving critical context for continuity.
 DEFAULT_SUMMARIZATION_PROMPT = """
-You are a summarization assistant.
-Your goal is to help main assistant to continue the conversation by creating an updated,
-concise summary integrating the previous summary (if any) with the new conversation history.
-Preserve ALL critical context needed for the main assistant
-to continue the task effectively. This includes key facts, decisions, tool usage
-results, and essential background. Do not omit details that would force the main
-assistant to re-gather information.
+You are a summarization assistant. Your goal is to help the main assistant continue a conversation by creating an updated, concise summary.
+You will integrate the previous summary (if any) with the new conversation history.
+
+It is CRITICAL to preserve the immediate context of the most recent exchange. Your summary MUST conclude with the user's last intent and the assistant's pending action.
+For example, if the user says "Yes" after the assistant asks "Do you want me to search?", the summary must explicitly state: "The user has confirmed that they want the assistant to proceed with the search for [topic]."
+
+Preserve ALL critical context:
+- The user's main, overarching goal.
+- The user's most recent, immediate intent.
+- Key decisions, facts, and entities (names, files, IDs).
+- Results from any tool usage.
+- Any pending questions or actions the assistant was about to take.
+
+Do not omit details that would force the main assistant to re-ask a question the user has already answered.
 Output *only* the updated summary text.
 """.strip()
 
 DEFAULT_CONTEXT_ENRICHMENT_PROMPT = """
-You are an information extraction assistant.
-Your goal is to help main assistant to continue the conversation by extracting
-important long-term/short-term informations.
-Analyze the conversation history and current context to extract key facts like
-user_name, user_roles, user_address, etc.
-Return only a JSON object containing a single key "response", whose value is
-another JSON object with these details (i.e., {"response": {"context_name": "value"}}).
+You are an information extraction assistant. Your goal is to help the main assistant by extracting important structured information from the conversation.
+Handle all data, especially personally identifiable information (PII), with strict confidentiality.
+
+Analyze the conversation to extract two types of information:
+1. **Stable Facts**: Key-value pairs that are unlikely to change often (e.g., user_name, user_id, project_name).
+2. **Conversational State**: The immediate task-related context. This is critical for maintaining continuity.
+   - "user_intent": The user's most recently stated goal (e.g., "find information about Albert Einstein").
+   - "pending_action": An action the assistant has proposed and is awaiting confirmation for or is about to execute (e.g., "search_internet").
+   - "action_parameters": A JSON object of parameters for the pending_action (e.g., {"query": "Albert Einstein"}).
+
+If an existing key needs to be updated (e.g., user changes their mind), replace the old value with the new one.
+Return only a JSON object containing a single key "response", whose value is another JSON object with these details (i.e., {"response": {"context_name": "value"}}).
 If no context can be extracted, return {"response": {}}.
 """.strip()
 

{zrb-1.8.6.dist-info → zrb-1.8.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: zrb
-Version: 1.8.6
+Version: 1.8.8
 Summary: Your Automation Powerhouse
 Home-page: https://github.com/state-alchemists/zrb
 License: AGPL-3.0-or-later
@@ -26,7 +26,7 @@ Requires-Dist: openai (>=1.76.0,<2.0.0) ; extra == "rag" or extra == "all"
 Requires-Dist: pdfplumber (>=0.11.6,<0.12.0) ; extra == "rag" or extra == "all"
 Requires-Dist: playwright (>=1.51.0,<2.0.0) ; extra == "playwright" or extra == "all"
 Requires-Dist: psutil (>=7.0.0,<8.0.0)
-Requires-Dist: pydantic-ai (>=0.2.7,<0.3.0)
+Requires-Dist: pydantic-ai (>=0.2.18,<0.3.0)
 Requires-Dist: pyjwt (>=2.10.1,<3.0.0)
 Requires-Dist: python-dotenv (>=1.1.0,<2.0.0)
 Requires-Dist: python-jose[cryptography] (>=3.4.0,<4.0.0)

{zrb-1.8.6.dist-info → zrb-1.8.8.dist-info}/RECORD

@@ -12,13 +12,13 @@ zrb/builtin/jwt.py,sha256=3M5uaQhJZbKQLjTUft1OwPz_JxtmK-xtkjxWjciOQho,2859
 zrb/builtin/llm/chat_session.py,sha256=ot2ss6yA4qIINg0nl3KJYnLag8H0eB9ggAgRGEUkZdE,6639
 zrb/builtin/llm/history.py,sha256=cnkOyO43uiMQ9cEvmqk-pPoCk1zCAH_fwAqSgBtsjzY,3079
 zrb/builtin/llm/input.py,sha256=Nw-26uTWp2QhUgKJcP_IMHmtk-b542CCSQ_vCOjhvhM,877
-zrb/builtin/llm/llm_ask.py,sha256=zXvXh4a2OPfvA3s4ZWVWafItLHRjDmaXoqsSHJXJxY4,4440
+zrb/builtin/llm/llm_ask.py,sha256=ltQlPEItcLW8pSZifUS5Ec9k8OvUo3__8W2MAG9J3IA,4438
 zrb/builtin/llm/previous-session.js,sha256=xMKZvJoAbrwiyHS0OoPrWuaKxWYLoyR5sguePIoCjTY,816
 zrb/builtin/llm/tool/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 zrb/builtin/llm/tool/api.py,sha256=yR9I0ZsI96OeQl9pgwORMASVuXsAL0a89D_iPS4C8Dc,1699
 zrb/builtin/llm/tool/cli.py,sha256=_CNEmEc6K2Z0i9ppYeM7jGpqaEdT3uxaWQatmxP3jKE,858
-zrb/builtin/llm/tool/code.py,sha256=X2wEHJlYUcJXhQUrpZ-WXr2BZFS6gXvbv18dEoeHzf4,7408
-zrb/builtin/llm/tool/file.py,sha256=-lw-_ts5UyrbfWL-LKGIelljfvU7D7MTL-C0jfNMxpY,17938
+zrb/builtin/llm/tool/code.py,sha256=7fozry3yKJ2793W4BQkiGnH4OgU0FCe87kNRG05DJZ4,7428
+zrb/builtin/llm/tool/file.py,sha256=hgm2s5zjGu2lo05Y0XisBu9HeaPPs2FyzwaoHBSFAhI,18355
 zrb/builtin/llm/tool/rag.py,sha256=yqx7vXXyrOCJjhQJl4s0TnLL-2uQUTuKRnkWlSQBW0M,7883
 zrb/builtin/llm/tool/sub_agent.py,sha256=_ItDE5MV_RZtnY_-IUsSMmm6mYaDY3YRINT0hVNsGkA,4702
 zrb/builtin/llm/tool/web.py,sha256=pXRLhcB_Y6z-2w4C4WezH8n-pg3PSMgt_bwn3aaqi6g,5479
@@ -217,7 +217,7 @@ zrb/callback/callback.py,sha256=PFhCqzfxdk6IAthmXcZ13DokT62xtBzJr_ciLw6I8Zg,4030
 zrb/cmd/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 zrb/cmd/cmd_result.py,sha256=L8bQJzWCpcYexIxHBNsXj2pT3BtLmWex0iJSMkvimOA,597
 zrb/cmd/cmd_val.py,sha256=7Doowyg6BK3ISSGBLt-PmlhzaEkBjWWm51cED6fAUOQ,1014
-zrb/config.py,sha256=FsDU_SyX3NL9xX3sMvaGflKjRyNm8k-x5OYSfR6dcFw,10211
+zrb/config.py,sha256=XJCJl3u6fo9OlL8V1HNLn5iKcSJjIICQk87CkBrA7IU,10222
 zrb/content_transformer/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 zrb/content_transformer/any_content_transformer.py,sha256=v8ZUbcix1GGeDQwB6OKX_1TjpY__ksxWVeqibwa_iZA,850
 zrb/content_transformer/content_transformer.py,sha256=STl77wW-I69QaGzCXjvkppngYFLufow8ybPLSyAvlHs,2404
@@ -246,7 +246,7 @@ zrb/input/option_input.py,sha256=TQB82ko5odgzkULEizBZi0e9TIHEbIgvdP0AR3RhA74,213
 zrb/input/password_input.py,sha256=szBojWxSP9QJecgsgA87OIYwQrY2AQ3USIKdDZY6snU,1465
 zrb/input/str_input.py,sha256=NevZHX9rf1g8eMatPyy-kUX3DglrVAQpzvVpKAzf7bA,81
 zrb/input/text_input.py,sha256=6T3MngWdUs0u0ZVs5Dl11w5KS7nN1RkgrIR_zKumzPM,3695
-zrb/llm_config.py,sha256=w_GSyQiJ_Q2jupej8xvQULScqPaSYICdNcH9J54W1lE,11696
+zrb/llm_config.py,sha256=9Pvm8SIW97C0C03tSZv-6aBsaFz4fiId3FG3znFteUk,13473
 zrb/llm_rate_limitter.py,sha256=RXdtPreMcmoYSE2Ab2StyHH95F0bD2pGmyySXs4gRio,3725
 zrb/runner/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 zrb/runner/cli.py,sha256=AbLTNqFy5FuyGQOWOjHZGaBC8e2yuE_Dx1sBdnisR18,6984
@@ -390,7 +390,7 @@ zrb/util/string/name.py,sha256=SXEfxJ1-tDOzHqmSV8kvepRVyMqs2XdV_vyoh_9XUu0,1584
 zrb/util/todo.py,sha256=VGISej2KQZERpornK-8X7bysp4JydMrMUTnG8B0-liI,20708
 zrb/xcom/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 zrb/xcom/xcom.py,sha256=o79rxR9wphnShrcIushA0Qt71d_p3ZTxjNf7x9hJB78,1571
-zrb-1.8.6.dist-info/METADATA,sha256=KZtv1CLg3gUa8fd4_ob39CMc-vhMFM8VW2tIiWTSqDs,9760
-zrb-1.8.6.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-zrb-1.8.6.dist-info/entry_points.txt,sha256=-Pg3ElWPfnaSM-XvXqCxEAa-wfVI6BEgcs386s8C8v8,46
-zrb-1.8.6.dist-info/RECORD,,
+zrb-1.8.8.dist-info/METADATA,sha256=x49O7t73u93ctMp-rTAz-MjelM8ZDyuVZ8vRFSEG54I,9761
+zrb-1.8.8.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+zrb-1.8.8.dist-info/entry_points.txt,sha256=-Pg3ElWPfnaSM-XvXqCxEAa-wfVI6BEgcs386s8C8v8,46
+zrb-1.8.8.dist-info/RECORD,,