cuga 0.1.0.dev1__tar.gz → 0.1.1.dev2__tar.gz

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cuga
-Version: 0.1.0.dev1
+Version: 0.1.1.dev2
 Summary: CUGA is an open-source generalist agent for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features.
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://cuga.dev

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "cuga"
-version = "0.1.0.dev1"
+version = "0.1.1.dev2"
 description = "CUGA is an open-source generalist agent for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features."
 readme = "README.md"
 license = "Apache-2.0"

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/cuga_graph/nodes/api/api_planner.py

@@ -1,7 +1,6 @@
 import json
 from typing import Literal
 
-from langgraph.graph.state import CompiledStateGraph
 
 from cuga.backend.activity_tracker.tracker import ActivityTracker, Step
 from cuga.backend.cuga_graph.nodes.api.variables_manager.manager import VariablesManager
@@ -19,13 +18,12 @@ from langgraph.types import Command
 from cuga.backend.cuga_graph.state.api_planner_history import HistoricalAction
 from loguru import logger
 
-from langgraph.prebuilt import create_react_agent
 from langchain_core.tools import tool
 
 from cuga.backend.llm.models import LLMManager
-from cuga.backend.llm.utils.helpers import load_one_prompt
 from cuga.config import settings
 from cuga.configurations.instructions_manager import InstructionsManager
+from cuga.backend.cuga_graph.nodes.api.tasks.reflection import reflection_task
 
 instructions_manager = InstructionsManager()
 var_manager = VariablesManager()
@@ -47,15 +45,7 @@ class ApiPlanner(BaseNode):
     def __init__(self, router_agent: APIPlannerAgent):
         super().__init__()
         self.name = router_agent.name
-        pmt_path = "reflection/prompts/system.jinja2"
-        pmt = load_one_prompt(pmt_path)
-        instructions = instructions_manager.get_instructions("api_reflection")
-        pmt_text = pmt.format(instructions=instructions)
-        self.guidance = create_react_agent(
-            model=llm_manager.get_model(settings.agent.planner.model),
-            tools=[think],
-            prompt=pmt_text,
-        )
+        self.guidance = reflection_task(llm=llm_manager.get_model(settings.agent.planner.model))
         self.agent = router_agent
         self.node = create_partial(
             ApiPlanner.node_handler,
@@ -71,7 +61,7 @@ class ApiPlanner(BaseNode):
 
     @staticmethod
     async def node_handler(
-        state: AgentState, agent: APIPlannerAgent, strategic_agent: CompiledStateGraph, name: str
+        state: AgentState, agent: APIPlannerAgent, strategic_agent, name: str
     ) -> Command[Literal['APICodePlannerAgent', 'ShortlisterAgent', 'PlanControllerAgent']]:
         # First time visit
         if (
@@ -81,15 +71,14 @@ class ApiPlanner(BaseNode):
         ):
             res_2 = await strategic_agent.ainvoke(
                 {
-                    "messages": [
-                        {
-                            "role": "user",
-                            "content": f"Variables history: {var_manager.get_variables_summary(last_n=5)}\n\ntask: {state.sub_task}\nHere's the history of actions/steps:\n\n{state.api_planner_history}\n\nUser information ( User already logged in ): {state.pi}\n\nCurrent datetime: {tracker.current_date}",
-                        }
-                    ]
+                    "instructions": instructions_manager.get_instructions("api_reflection"),
+                    "current_task": state.sub_task,
+                    "agent_history": str(state.api_planner_history),
+                    "shortlister_agent_output": "N/A",  # This would need to be populated from actual shortlister output
+                    "coder_agent_output": f"Variables history: {var_manager.get_variables_summary(last_n=5)}\n\nUser information ( User already logged in ): {state.pi}\n\nCurrent datetime: {tracker.current_date}",
                 }
             )
-            summary = res_2['messages'][-1].content
+            summary = res_2.content
             state.guidance = summary
             tracker.collect_step(step=Step(name=name, data=summary))
             logger.debug(f"Guidance:\n{summary}")

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/prompts/system.jinja2

@@ -1,6 +1,6 @@
 You are an expert AI assistant responsible for selecting relevant APIs to fulfill a user's request.
 Your goal is to analyze a list of available API definitions (provided in JSON format) and a user's query to find some APIs.
-Based on this analysis, you must identify the APIs that are most relevant to achieve the user's goal. **Your final output must list at least 4 APIs in the `result` field.** These APIs should be ranked by their `relevance_score` from highest to lowest. If you find fewer than 4 APIs that are strongly relevant, you must include APIs with lower or marginal relevance to meet the count of at least 4. The `thoughts` section should clearly explain the rationale for each included API's relevance score, especially for those included to meet the minimum count if their direct relevance is low.
+Based on this analysis, you must identify the APIs that are most relevant to achieve the user's goal. **Your final output must list at least 1 APIs in the `result` field.** These APIs should be ranked by their `relevance_score` from highest to lowest. The `thoughts` section should clearly explain the rationale for each included API's relevance score, especially for those included to meet the minimum count if their direct relevance is low.
 
 Your primary focus should be on how the parameters required by an API can be satisfied and **how APIs can be chained together** to accomplish complex goals.
 Consider the following rules for parameter sourcing:
@@ -183,7 +183,7 @@ You need to evaluate each API's relevance based on:
         {
             "name": "pet_management_delete_pet",
             "relevance_score": 0.05,
-            "reasoning": "Low relevance as user wants to update, not delete pets. Included to meet minimum of 4 APIs. Key parameter: 'petId' (could be sourced from other APIs' responses). Response schema returns simple message object but functionality doesn't align with user's query."
+            "reasoning": "Low relevance as user wants to update, not delete pets. Included to meet minimum of 1 APIs. Key parameter: 'petId' (could be sourced from other APIs' responses). Response schema returns simple message object but functionality doesn't align with user's query."
         }
     ]
 }

cuga-0.1.0.dev1/src/cuga/backend/cuga_graph/nodes/api/reflection/prompts/system.jinja2 → cuga-0.1.1.dev2/src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/reflection_system.jinja2

@@ -53,4 +53,11 @@ After you have used the `think` tool for your internal deliberation, you must th
 2.  **Summary of Progress** made so far on the **Current Task**.
 3.  **Strategic Recommendation** for the next steps or confirmation of task completion for the **Current Task**.
 
-Your turn: Analyze the provided current task, history, and outputs, then use the `think` tool for your internal process, and finally, present your consolidated guidance and summary as the final output.
+Your turn: Analyze the provided current task, history, and outputs, then use the `think` tool for your internal process, and finally, present your consolidated guidance and summary as the final output.
+
+{% if instructions -%}
+
+## Special Instructions
+{{ instructions }}
+
+{%- endif %}

cuga-0.1.1.dev2/src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/reflection_user.jinja2

@@ -0,0 +1,7 @@
+**Current Task**: {{current_task}}
+
+**Agent History**: {{agent_history}}
+
+**Shortlister Agent Output**: {{shortlister_agent_output}}
+
+**Coder Agent Output**: {{coder_agent_output}}

cuga-0.1.1.dev2/src/cuga/backend/cuga_graph/nodes/api/tasks/reflection.py

@@ -0,0 +1,14 @@
+from langchain_core.runnables import Runnable
+from cuga.backend.cuga_graph.nodes.shared.base_agent import BaseAgent
+from cuga.backend.llm.models import LLMManager
+from cuga.backend.llm.utils.helpers import load_prompt_simple
+
+llm_manager = LLMManager()
+
+
+def reflection_task(llm, enable_format=False) -> Runnable:
+    prompt_template = load_prompt_simple(
+        "./prompts/reflection_system.jinja2",
+        "./prompts/reflection_user.jinja2",
+    )
+    return BaseAgent.get_chain(prompt_template, llm, wx_json_mode="no_format")

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/cuga_graph/nodes/shared/base_agent.py

@@ -17,6 +17,7 @@ from loguru import logger
 
 from cuga.backend.cuga_graph.nodes.api.api_planner_agent.prompts.load_prompt import (
     APIPlannerOutput,
+    APIPlannerOutputLite,
     APIPlannerOutputWX,
 )
 
@@ -77,7 +78,7 @@ JSON schema:
         #     return prompt_template | llm.bind(extra_body={"guided_json": schema.model_json_schema()}) | parser
         if isinstance(llm, ChatWatsonx):
             logger.debug("Loading LLM for watsonx")
-            if schema == APIPlannerOutput:
+            if schema == APIPlannerOutput or schema == APIPlannerOutputLite:
                 logger.debug("Switched to watsonx schema... for APIPlannerOutput")
                 schema = APIPlannerOutputWX
             parser = PydanticOutputParser(pydantic_object=schema)

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/cuga_graph/utils/nodes_names.py

@@ -12,7 +12,7 @@ class NodeNames:
     REUSE_AGENT = "ReuseAgent"
     API_CODE_PLANNER_AGENT = "APICodePlannerAgent"
     SHORTLISTER_AGENT = "ShortlisterAgent"
-    DECOMPOSITION_AGENT = "DecompositionAgent"
+    DECOMPOSITION_AGENT = "TaskDecompositionAgent"
     WAIT_FOR_RESPONSE = "WaitForResponse"
     CHAT_AGENT = "ChatAgent"
     API_PLANNER_AGENT = "APIPlannerAgent"

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/server/main.py

@@ -448,9 +448,10 @@ async def event_stream(query: str, api_mode=False, resume=None):
                     )
                     name = ((event.split("\n")[0]).split(":")[1]).strip()
                     logger.debug("Yield {}".format(event))
-                    yield StreamEvent(name=name, data=event).format(
-                        app_state.output_format, thread_id=app_state.thread_id
-                    )
+                    if name not in ["ChatAgent"]:
+                        yield StreamEvent(name=name, data=event).format(
+                            app_state.output_format, thread_id=app_state.thread_id
+                        )
     except Exception as e:
         logger.exception(e)
         app_state.tracker.finish_task(

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/tools_env/code_sandbox/sandbox.py

@@ -89,15 +89,15 @@ def get_premable(is_local=False, current_date=None):
     )
 
     # Check if structured tools should be enabled
-    if not settings.features.local_sandbox and tracker.tools is not None and len(tracker.tools) > 0:
-        logger.warning("Structured tools not enabled - local sandbox mode is active")
-        tool_import_code = ""
-        tool_init_code = ""
-        tool_invocation_code = ""
-    else:
+    if settings.features.local_sandbox and tracker.tools is not None and len(tracker.tools) > 0:
         tool_import_code = structured_tools_import
         tool_init_code = structured_tools_init
         tool_invocation_code = structured_tools_invocation
+    else:
+        logger.warning("Structured tools not enabled")
+        tool_import_code = ""
+        tool_init_code = ""
+        tool_invocation_code = ""
 
     preamble = (
         """

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/backend/tools_env/registry/utils/api_utils.py

@@ -28,6 +28,9 @@ async def get_apis(app_name: str):
     try:
         logger.debug("calling get_apis")
         tools = tracker.get_tools_by_server(app_name)
+        if not settings.advanced_features.registry:
+            logger.debug("Registry is not enabled, using external tools")
+            return tools
         if tools:
             return tools
     except Exception as e:
@@ -76,6 +79,9 @@ async def get_apps() -> List[AppDefinition]:
     url = f'http://127.0.0.1:{settings.server_ports.registry}/applications'
     headers = {'accept': 'application/json'}
     external_apps = tracker.apps
+    if not settings.advanced_features.registry:
+        logger.debug("Registry is not enabled, using external apps")
+        return external_apps
     logger.debug(f"External apps are {external_apps}")
     try:
         async with aiohttp.ClientSession() as session:

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/config.py

@@ -8,7 +8,7 @@ from pathlib import Path
 
 # third-party imports
 import dynaconf
-from dotenv import load_dotenv
+from dotenv import load_dotenv, find_dotenv
 from dynaconf import Dynaconf, Validator
 from loguru import logger
 
@@ -32,21 +32,24 @@ MODES_DIR = os.path.join(CONFIGURATIONS_DIR, "modes")
 
 # from feature_flags import FeatureFlags as flags
 
-# Try CI/CD environment variable first
-env_path = os.getenv("ENV_FILE")
+# 1) Let users (or CI) force a path when needed
+if os.getenv("ENV_FILE"):
+    load_dotenv(os.getenv("ENV_FILE"), override=True)
+else:
+    # 2) Try to find it when working from a git clone (searches up from CWD)
+    path = find_dotenv(filename=".env", usecwd=True)
+    if not path:
+        # 3) Try again when your code is used as an installed package
+        #    (searches up from the module’s location)
+        path = find_dotenv(filename=".env", usecwd=False)
 
-if not env_path:
-    # Fallback when running from an installed wheel or from source
-    env_path = ENV_FILE_PATH
-
-load_dotenv(dotenv_path=env_path, override=True)
+    load_dotenv(path, override=False)
 
 for key, value in os.environ.items():
     if key.startswith("WA_"):
         new_key = key[3:]
         os.environ[new_key] = value
 
-
 app_mapping = {
     urlparse(os.getenv("WA_REDDIT")).netloc: "reddit",
     urlparse(os.getenv("WA_SHOPPING")).netloc: "shopping",
@@ -89,6 +92,7 @@ validators = [
     Validator("features.forced_apps", default=None),
     Validator("features.thoughts", default=True),
     Validator("features.code_generation", default="accurate"),
+    Validator("advanced_features.registry", default=True),
     Validator("features.task_decomposition", default=False),
     Validator("advanced_features.langfuse_tracing", default=False),
     Validator("advanced_features.benchmark", default="default"),

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/configurations/instructions/instructions.toml

@@ -19,5 +19,8 @@ instructions = "@format {this.base_path}/{this.instruction_set}/code_agent.md"
 [api_reflection]
 instructions = "@format {this.base_path}/{this.instruction_set}/reflection.md"
 
+[api_code_planner]
+instructions = "@format {this.base_path}/{this.instruction_set}/api_code_planner.md"
+
 [answer]
 instructions = "@format {this.base_path}/{this.instruction_set}/answer.md"

cuga-0.1.1.dev2/src/cuga/configurations/modes/accurate.toml

@@ -0,0 +1,11 @@
+[features]
+chat = false
+task_decomposition = true
+local_sandbox = true
+thoughts = true
+code_output_summary = true
+code_generation = "accurate"
+code_output_reflection = true
+save_reuse = false
+forced_apps = []
+final_answer = true

cuga-0.1.0.dev1/src/cuga/configurations/modes/accurate.toml → cuga-0.1.1.dev2/src/cuga/configurations/modes/balanced.toml

@@ -8,4 +8,4 @@ code_generation = "fast"
 code_output_reflection = true
 save_reuse = false
 forced_apps = []
-final_answer = true
+final_answer = true

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga/settings.toml

@@ -9,7 +9,7 @@ max_steps = 55
 start_url = "https://opensource-demo.orangehrmlive.com/web/index.php/auth/login"
 
 [features]
-cuga_mode = "accurate"
+cuga_mode = "balanced"
 
 [advanced_features]
 web_arena_eval = false

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cuga
-Version: 0.1.0.dev1
+Version: 0.1.1.dev2
 Summary: CUGA is an open-source generalist agent for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features.
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://cuga.dev

{cuga-0.1.0.dev1 → cuga-0.1.1.dev2}/src/cuga.egg-info/SOURCES.txt

@@ -110,7 +110,6 @@ src/cuga/backend/cuga_graph/nodes/api/code_agent/code_agent.py
 src/cuga/backend/cuga_graph/nodes/api/code_agent/model.py
 src/cuga/backend/cuga_graph/nodes/api/code_agent/prompts/system_accurate.jinja2
 src/cuga/backend/cuga_graph/nodes/api/code_agent/prompts/system_fast.jinja2
-src/cuga/backend/cuga_graph/nodes/api/reflection/prompts/system.jinja2
 src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/__init__.py
 src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/shortlister_agent.py
 src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/prompts/__init__.py
@@ -120,7 +119,10 @@ src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/prompts/system_parameter
 src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/prompts/user.jinja2
 src/cuga/backend/cuga_graph/nodes/api/shortlister_agent/prompts/user_parameters.jinja2
 src/cuga/backend/cuga_graph/nodes/api/tasks/__init__.py
+src/cuga/backend/cuga_graph/nodes/api/tasks/reflection.py
 src/cuga/backend/cuga_graph/nodes/api/tasks/summarize_code.py
+src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/reflection_system.jinja2
+src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/reflection_user.jinja2
 src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/summary_system.jinja2
 src/cuga/backend/cuga_graph/nodes/api/tasks/prompts/summary_user.jinja2
 src/cuga/backend/cuga_graph/nodes/api/variables_manager/__init__.py
@@ -308,6 +310,7 @@ src/cuga/configurations/instructions_manager.py
 src/cuga/configurations/set_from_one_file.py
 src/cuga/configurations/instructions/instructions.toml
 src/cuga/configurations/instructions/default/answer.md
+src/cuga/configurations/instructions/default/api_code_planner.md
 src/cuga/configurations/instructions/default/api_planner.md
 src/cuga/configurations/instructions/default/code_agent.md
 src/cuga/configurations/instructions/default/plan_controller.md
@@ -319,6 +322,7 @@ src/cuga/configurations/models/settings.openai.toml
 src/cuga/configurations/models/settings.rits.toml
 src/cuga/configurations/models/settings.watsonx.toml
 src/cuga/configurations/modes/accurate.toml
+src/cuga/configurations/modes/balanced.toml
 src/cuga/configurations/modes/custom.toml
 src/cuga/configurations/modes/fast.toml
 src/cuga/configurations/modes/save_reuse_fast.toml