将flask改成fastapi

agent/component/__init__.py

@@ -0,0 +1,58 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import os
+import importlib
+import inspect
+from types import ModuleType
+from typing import Dict, Type
+
+_package_path = os.path.dirname(__file__)
+__all_classes: Dict[str, Type] = {}
+
+def _import_submodules() -> None:
+    for filename in os.listdir(_package_path): # noqa: F821
+        if filename.startswith("__") or not filename.endswith(".py") or filename.startswith("base"):
+            continue
+        module_name = filename[:-3]
+
+        try:
+            module = importlib.import_module(f".{module_name}", package=__name__)
+            _extract_classes_from_module(module)  # noqa: F821
+        except ImportError as e:
+            print(f"Warning: Failed to import module {module_name}: {str(e)}")
+
+def _extract_classes_from_module(module: ModuleType) -> None:
+    for name, obj in inspect.getmembers(module):
+        if (inspect.isclass(obj) and
+                obj.__module__ == module.__name__ and not name.startswith("_")):
+            __all_classes[name] = obj
+            globals()[name] = obj
+
+_import_submodules()
+
+__all__ = list(__all_classes.keys()) + ["__all_classes"]
+
+del _package_path, _import_submodules, _extract_classes_from_module
+
+
+def component_class(class_name):
+    for mdl in ["agent.component", "agent.tools", "rag.flow"]:
+        try:
+            return getattr(importlib.import_module(mdl), class_name)
+        except Exception:
+            pass
+    assert False, f"Can't import {class_name}"

agent/component/agent_with_tools.py

@@ -0,0 +1,352 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import logging
+import os
+import re
+from concurrent.futures import ThreadPoolExecutor
+from copy import deepcopy
+from functools import partial
+from typing import Any
+
+import json_repair
+from timeit import default_timer as timer
+from agent.tools.base import LLMToolPluginCallSession, ToolParamBase, ToolBase, ToolMeta
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
+from api.db.services.mcp_server_service import MCPServerService
+from api.utils.api_utils import timeout
+from rag.prompts.generator import next_step, COMPLETE_TASK, analyze_task, \
+    citation_prompt, reflect, rank_memories, kb_prompt, citation_plus, full_question, message_fit_in
+from rag.utils.mcp_tool_call_conn import MCPToolCallSession, mcp_tool_metadata_to_openai_tool
+from agent.component.llm import LLMParam, LLM
+
+
+class AgentParam(LLMParam, ToolParamBase):
+    """
+    Define the Agent component parameters.
+    """
+
+    def __init__(self):
+        self.meta:ToolMeta = {
+                "name": "agent",
+                "description": "This is an agent for a specific task.",
+                "parameters": {
+                    "user_prompt": {
+                        "type": "string",
+                        "description": "This is the order you need to send to the agent.",
+                        "default": "",
+                        "required": True
+                    },
+                    "reasoning": {
+                        "type": "string",
+                        "description": (
+                            "Supervisor's reasoning for choosing the this agent. "
+                            "Explain why this agent is being invoked and what is expected of it."
+                        ),
+                        "required": True
+                    },
+                    "context": {
+                        "type": "string",
+                        "description": (
+                                "All relevant background information, prior facts, decisions, "
+                                "and state needed by the agent to solve the current query. "
+                                "Should be as detailed and self-contained as possible."
+                            ),
+                        "required": True
+                    },
+                }
+            }
+        super().__init__()
+        self.function_name = "agent"
+        self.tools = []
+        self.mcp = []
+        self.max_rounds = 5
+        self.description = ""
+
+
+class Agent(LLM, ToolBase):
+    component_name = "Agent"
+
+    def __init__(self, canvas, id, param: LLMParam):
+        LLM.__init__(self, canvas, id, param)
+        self.tools = {}
+        for cpn in self._param.tools:
+            cpn = self._load_tool_obj(cpn)
+            self.tools[cpn.get_meta()["function"]["name"]] = cpn
+
+        self.chat_mdl = LLMBundle(self._canvas.get_tenant_id(), TenantLLMService.llm_id2llm_type(self._param.llm_id), self._param.llm_id,
+                                  max_retries=self._param.max_retries,
+                                  retry_interval=self._param.delay_after_error,
+                                  max_rounds=self._param.max_rounds,
+                                  verbose_tool_use=True
+                                  )
+        self.tool_meta = [v.get_meta() for _,v in self.tools.items()]
+
+        for mcp in self._param.mcp:
+            _, mcp_server = MCPServerService.get_by_id(mcp["mcp_id"])
+            tool_call_session = MCPToolCallSession(mcp_server, mcp_server.variables)
+            for tnm, meta in mcp["tools"].items():
+                self.tool_meta.append(mcp_tool_metadata_to_openai_tool(meta))
+                self.tools[tnm] = tool_call_session
+        self.callback = partial(self._canvas.tool_use_callback, id)
+        self.toolcall_session = LLMToolPluginCallSession(self.tools, self.callback)
+        #self.chat_mdl.bind_tools(self.toolcall_session, self.tool_metas)
+
+    def _load_tool_obj(self, cpn: dict) -> object:
+        from agent.component import component_class
+        param = component_class(cpn["component_name"] + "Param")()
+        param.update(cpn["params"])
+        try:
+            param.check()
+        except Exception as e:
+            self.set_output("_ERROR", cpn["component_name"] + f" configuration error: {e}")
+            raise
+        cpn_id = f"{self._id}-->" + cpn.get("name", "").replace(" ", "_")
+        return component_class(cpn["component_name"])(self._canvas, cpn_id, param)
+
+    def get_meta(self) -> dict[str, Any]:
+        self._param.function_name= self._id.split("-->")[-1]
+        m = super().get_meta()
+        if hasattr(self._param, "user_prompt") and self._param.user_prompt:
+            m["function"]["parameters"]["properties"]["user_prompt"] = self._param.user_prompt
+        return m
+
+    def get_input_form(self) -> dict[str, dict]:
+        res = {}
+        for k, v in self.get_input_elements().items():
+            res[k] = {
+                "type": "line",
+                "name": v["name"]
+            }
+        for cpn in self._param.tools:
+            if not isinstance(cpn, LLM):
+                continue
+            res.update(cpn.get_input_form())
+        return res
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 20*60)))
+    def _invoke(self, **kwargs):
+        if kwargs.get("user_prompt"):
+            usr_pmt = ""
+            if kwargs.get("reasoning"):
+                usr_pmt += "\nREASONING:\n{}\n".format(kwargs["reasoning"])
+            if kwargs.get("context"):
+                usr_pmt += "\nCONTEXT:\n{}\n".format(kwargs["context"])
+            if usr_pmt:
+                usr_pmt += "\nQUERY:\n{}\n".format(str(kwargs["user_prompt"]))
+            else:
+                usr_pmt = str(kwargs["user_prompt"])
+            self._param.prompts = [{"role": "user", "content": usr_pmt}]
+
+        if not self.tools:
+            return LLM._invoke(self, **kwargs)
+
+        prompt, msg, user_defined_prompt = self._prepare_prompt_variables()
+
+        downstreams = self._canvas.get_component(self._id)["downstream"] if self._canvas.get_component(self._id) else []
+        ex = self.exception_handler()
+        if any([self._canvas.get_component_obj(cid).component_name.lower()=="message" for cid in downstreams]) and not self._param.output_structure and not (ex and ex["goto"]):
+            self.set_output("content", partial(self.stream_output_with_tools, prompt, msg, user_defined_prompt))
+            return
+
+        _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
+        use_tools = []
+        ans = ""
+        for delta_ans, tk in self._react_with_tools_streamly(prompt, msg, use_tools, user_defined_prompt):
+            ans += delta_ans
+
+        if ans.find("**ERROR**") >= 0:
+            logging.error(f"Agent._chat got error. response: {ans}")
+            if self.get_exception_default_value():
+                self.set_output("content", self.get_exception_default_value())
+            else:
+                self.set_output("_ERROR", ans)
+            return
+
+        self.set_output("content", ans)
+        if use_tools:
+            self.set_output("use_tools", use_tools)
+        return ans
+
+    def stream_output_with_tools(self, prompt, msg, user_defined_prompt={}):
+        _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
+        answer_without_toolcall = ""
+        use_tools = []
+        for delta_ans,_ in self._react_with_tools_streamly(prompt, msg, use_tools, user_defined_prompt):
+            if delta_ans.find("**ERROR**") >= 0:
+                if self.get_exception_default_value():
+                    self.set_output("content", self.get_exception_default_value())
+                    yield self.get_exception_default_value()
+                else:
+                    self.set_output("_ERROR", delta_ans)
+            answer_without_toolcall += delta_ans
+            yield delta_ans
+
+        self.set_output("content", answer_without_toolcall)
+        if use_tools:
+            self.set_output("use_tools", use_tools)
+
+    def _gen_citations(self, text):
+        retrievals = self._canvas.get_reference()
+        retrievals = {"chunks": list(retrievals["chunks"].values()), "doc_aggs": list(retrievals["doc_aggs"].values())}
+        formated_refer = kb_prompt(retrievals, self.chat_mdl.max_length, True)
+        for delta_ans in self._generate_streamly([{"role": "system", "content": citation_plus("\n\n".join(formated_refer))},
+                                                  {"role": "user", "content": text}
+                                                  ]):
+            yield delta_ans
+
+    def _react_with_tools_streamly(self, prompt, history: list[dict], use_tools, user_defined_prompt={}):
+        token_count = 0
+        tool_metas = self.tool_meta
+        hist = deepcopy(history)
+        last_calling = ""
+        if len(hist) > 3:
+            st = timer()
+            user_request = full_question(messages=history, chat_mdl=self.chat_mdl)
+            self.callback("Multi-turn conversation optimization", {}, user_request, elapsed_time=timer()-st)
+        else:
+            user_request = history[-1]["content"]
+
+        def use_tool(name, args):
+            nonlocal hist, use_tools, token_count,last_calling,user_request
+            logging.info(f"{last_calling=} == {name=}")
+            # Summarize of function calling
+            #if all([
+            #    isinstance(self.toolcall_session.get_tool_obj(name), Agent),
+            #    last_calling,
+            #    last_calling != name
+            #]):
+            #    self.toolcall_session.get_tool_obj(name).add2system_prompt(f"The chat history with other agents are as following: \n" + self.get_useful_memory(user_request, str(args["user_prompt"]),user_defined_prompt))
+            last_calling = name
+            tool_response = self.toolcall_session.tool_call(name, args)
+            use_tools.append({
+                "name": name,
+                "arguments": args,
+                "results": tool_response
+            })
+            # self.callback("add_memory", {}, "...")
+            #self.add_memory(hist[-2]["content"], hist[-1]["content"], name, args, str(tool_response), user_defined_prompt)
+
+            return name, tool_response
+
+        def complete():
+            nonlocal hist
+            need2cite = self._param.cite and self._canvas.get_reference()["chunks"] and self._id.find("-->") < 0
+            cited = False
+            if hist[0]["role"] == "system" and need2cite:
+                if len(hist) < 7:
+                    hist[0]["content"] += citation_prompt()
+                    cited = True
+            yield "", token_count
+
+            _hist = hist
+            if len(hist) > 12:
+                _hist = [hist[0], hist[1], *hist[-10:]]
+            entire_txt = ""
+            for delta_ans in self._generate_streamly(_hist):
+                if not need2cite or cited:
+                    yield delta_ans, 0
+                entire_txt += delta_ans
+            if not need2cite or cited:
+                return
+
+            st = timer()
+            txt = ""
+            for delta_ans in self._gen_citations(entire_txt):
+                yield delta_ans, 0
+                txt += delta_ans
+
+            self.callback("gen_citations", {}, txt, elapsed_time=timer()-st)
+
+        def append_user_content(hist, content):
+            if hist[-1]["role"] == "user":
+                hist[-1]["content"] += content
+            else:
+                hist.append({"role": "user", "content": content})
+
+        st = timer()
+        task_desc = analyze_task(self.chat_mdl, prompt, user_request, tool_metas, user_defined_prompt)
+        self.callback("analyze_task", {}, task_desc, elapsed_time=timer()-st)
+        for _ in range(self._param.max_rounds + 1):
+            response, tk = next_step(self.chat_mdl, hist, tool_metas, task_desc, user_defined_prompt)
+            # self.callback("next_step", {}, str(response)[:256]+"...")
+            token_count += tk
+            hist.append({"role": "assistant", "content": response})
+            try:
+                functions = json_repair.loads(re.sub(r"```.*", "", response))
+                if not isinstance(functions, list):
+                    raise TypeError(f"List should be returned, but `{functions}`")
+                for f in functions:
+                    if not isinstance(f, dict):
+                        raise TypeError(f"An object type should be returned, but `{f}`")
+                with ThreadPoolExecutor(max_workers=5) as executor:
+                    thr = []
+                    for func in functions:
+                        name = func["name"]
+                        args = func["arguments"]
+                        if name == COMPLETE_TASK:
+                            append_user_content(hist, f"Respond with a formal answer. FORGET(DO NOT mention) about `{COMPLETE_TASK}`. The language for the response MUST be as the same as the first user request.\n")
+                            for txt, tkcnt in complete():
+                                yield txt, tkcnt
+                            return
+
+                        thr.append(executor.submit(use_tool, name, args))
+
+                    st = timer()
+                    reflection = reflect(self.chat_mdl, hist, [th.result() for th in thr], user_defined_prompt)
+                    append_user_content(hist, reflection)
+                    self.callback("reflection", {}, str(reflection), elapsed_time=timer()-st)
+
+            except Exception as e:
+                logging.exception(msg=f"Wrong JSON argument format in LLM ReAct response: {e}")
+                e = f"\nTool call error, please correct the input parameter of response format and call it again.\n *** Exception ***\n{e}"
+                append_user_content(hist, str(e))
+
+        logging.warning( f"Exceed max rounds: {self._param.max_rounds}")
+        final_instruction = f"""
+{user_request}
+IMPORTANT: You have reached the conversation limit. Based on ALL the information and research you have gathered so far, please provide a DIRECT and COMPREHENSIVE final answer to the original request.
+Instructions:
+1. SYNTHESIZE all information collected during this conversation
+2. Provide a COMPLETE response using existing data - do not suggest additional research
+3. Structure your response as a FINAL DELIVERABLE, not a plan
+4. If information is incomplete, state what you found and provide the best analysis possible with available data
+5. DO NOT mention conversation limits or suggest further steps
+6. Focus on delivering VALUE with the information already gathered
+Respond immediately with your final comprehensive answer.
+        """
+        append_user_content(hist, final_instruction)
+
+        for txt, tkcnt in complete():
+            yield txt, tkcnt
+
+    def get_useful_memory(self, goal: str, sub_goal:str, topn=3, user_defined_prompt:dict={}) -> str:
+        # self.callback("get_useful_memory", {"topn": 3}, "...")
+        mems = self._canvas.get_memory()
+        rank = rank_memories(self.chat_mdl, goal, sub_goal, [summ for (user, assist, summ) in mems], user_defined_prompt)
+        try:
+            rank = json_repair.loads(re.sub(r"```.*", "", rank))[:topn]
+            mems = [mems[r] for r in rank]
+            return "\n\n".join([f"User: {u}\nAgent: {a}" for u, a,_ in mems])
+        except Exception as e:
+            logging.exception(e)
+
+        return "Error occurred."
+
+    def reset(self):
+        for k, cpn in self.tools.items():
+            cpn.reset()
+

agent/component/base.py

@@ -0,0 +1,564 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import re
+import time
+from abc import ABC
+import builtins
+import json
+import os
+import logging
+from typing import Any, List, Union
+import pandas as pd
+import trio
+from agent import settings
+from api.utils.api_utils import timeout
+
+
+_FEEDED_DEPRECATED_PARAMS = "_feeded_deprecated_params"
+_DEPRECATED_PARAMS = "_deprecated_params"
+_USER_FEEDED_PARAMS = "_user_feeded_params"
+_IS_RAW_CONF = "_is_raw_conf"
+
+
+class ComponentParamBase(ABC):
+    def __init__(self):
+        self.message_history_window_size = 13
+        self.inputs = {}
+        self.outputs = {}
+        self.description = ""
+        self.max_retries = 0
+        self.delay_after_error = 2.0
+        self.exception_method = None
+        self.exception_default_value = None
+        self.exception_goto = None
+        self.debug_inputs = {}
+
+    def set_name(self, name: str):
+        self._name = name
+        return self
+
+    def check(self):
+        raise NotImplementedError("Parameter Object should be checked.")
+
+    @classmethod
+    def _get_or_init_deprecated_params_set(cls):
+        if not hasattr(cls, _DEPRECATED_PARAMS):
+            setattr(cls, _DEPRECATED_PARAMS, set())
+        return getattr(cls, _DEPRECATED_PARAMS)
+
+    def _get_or_init_feeded_deprecated_params_set(self, conf=None):
+        if not hasattr(self, _FEEDED_DEPRECATED_PARAMS):
+            if conf is None:
+                setattr(self, _FEEDED_DEPRECATED_PARAMS, set())
+            else:
+                setattr(
+                    self,
+                    _FEEDED_DEPRECATED_PARAMS,
+                    set(conf[_FEEDED_DEPRECATED_PARAMS]),
+                )
+        return getattr(self, _FEEDED_DEPRECATED_PARAMS)
+
+    def _get_or_init_user_feeded_params_set(self, conf=None):
+        if not hasattr(self, _USER_FEEDED_PARAMS):
+            if conf is None:
+                setattr(self, _USER_FEEDED_PARAMS, set())
+            else:
+                setattr(self, _USER_FEEDED_PARAMS, set(conf[_USER_FEEDED_PARAMS]))
+        return getattr(self, _USER_FEEDED_PARAMS)
+
+    def get_user_feeded(self):
+        return self._get_or_init_user_feeded_params_set()
+
+    def get_feeded_deprecated_params(self):
+        return self._get_or_init_feeded_deprecated_params_set()
+
+    @property
+    def _deprecated_params_set(self):
+        return {name: True for name in self.get_feeded_deprecated_params()}
+
+    def __str__(self):
+        return json.dumps(self.as_dict(), ensure_ascii=False)
+
+    def as_dict(self):
+        def _recursive_convert_obj_to_dict(obj):
+            ret_dict = {}
+            if isinstance(obj, dict):
+                for k,v in obj.items():
+                    if isinstance(v, dict) or (v and type(v).__name__ not in dir(builtins)):
+                        ret_dict[k] = _recursive_convert_obj_to_dict(v)
+                    else:
+                        ret_dict[k] = v
+                return ret_dict
+
+            for attr_name in list(obj.__dict__):
+                if attr_name in [_FEEDED_DEPRECATED_PARAMS, _DEPRECATED_PARAMS, _USER_FEEDED_PARAMS, _IS_RAW_CONF]:
+                    continue
+                # get attr
+                attr = getattr(obj, attr_name)
+                if isinstance(attr, pd.DataFrame):
+                    ret_dict[attr_name] = attr.to_dict()
+                    continue
+                if isinstance(attr, dict) or (attr and type(attr).__name__ not in dir(builtins)):
+                    ret_dict[attr_name] = _recursive_convert_obj_to_dict(attr)
+                else:
+                    ret_dict[attr_name] = attr
+
+            return ret_dict
+
+        return _recursive_convert_obj_to_dict(self)
+
+    def update(self, conf, allow_redundant=False):
+        update_from_raw_conf = conf.get(_IS_RAW_CONF, True)
+        if update_from_raw_conf:
+            deprecated_params_set = self._get_or_init_deprecated_params_set()
+            feeded_deprecated_params_set = (
+                self._get_or_init_feeded_deprecated_params_set()
+            )
+            user_feeded_params_set = self._get_or_init_user_feeded_params_set()
+            setattr(self, _IS_RAW_CONF, False)
+        else:
+            feeded_deprecated_params_set = (
+                self._get_or_init_feeded_deprecated_params_set(conf)
+            )
+            user_feeded_params_set = self._get_or_init_user_feeded_params_set(conf)
+
+        def _recursive_update_param(param, config, depth, prefix):
+            if depth > settings.PARAM_MAXDEPTH:
+                raise ValueError("Param define nesting too deep!!!, can not parse it")
+
+            inst_variables = param.__dict__
+            redundant_attrs = []
+            for config_key, config_value in config.items():
+                # redundant attr
+                if config_key not in inst_variables:
+                    if not update_from_raw_conf and config_key.startswith("_"):
+                        setattr(param, config_key, config_value)
+                    else:
+                        setattr(param, config_key, config_value)
+                        # redundant_attrs.append(config_key)
+                    continue
+
+                full_config_key = f"{prefix}{config_key}"
+
+                if update_from_raw_conf:
+                    # add user feeded params
+                    user_feeded_params_set.add(full_config_key)
+
+                    # update user feeded deprecated param set
+                    if full_config_key in deprecated_params_set:
+                        feeded_deprecated_params_set.add(full_config_key)
+
+                # supported attr
+                attr = getattr(param, config_key)
+                if type(attr).__name__ in dir(builtins) or attr is None:
+                    setattr(param, config_key, config_value)
+
+                else:
+                    # recursive set obj attr
+                    sub_params = _recursive_update_param(
+                        attr, config_value, depth + 1, prefix=f"{prefix}{config_key}."
+                    )
+                    setattr(param, config_key, sub_params)
+
+            if not allow_redundant and redundant_attrs:
+                raise ValueError(
+                    f"cpn `{getattr(self, '_name', type(self))}` has redundant parameters: `{[redundant_attrs]}`"
+                )
+
+            return param
+
+        return _recursive_update_param(param=self, config=conf, depth=0, prefix="")
+
+    def extract_not_builtin(self):
+        def _get_not_builtin_types(obj):
+            ret_dict = {}
+            for variable in obj.__dict__:
+                attr = getattr(obj, variable)
+                if attr and type(attr).__name__ not in dir(builtins):
+                    ret_dict[variable] = _get_not_builtin_types(attr)
+
+            return ret_dict
+
+        return _get_not_builtin_types(self)
+
+    def validate(self):
+        self.builtin_types = dir(builtins)
+        self.func = {
+            "ge": self._greater_equal_than,
+            "le": self._less_equal_than,
+            "in": self._in,
+            "not_in": self._not_in,
+            "range": self._range,
+        }
+        home_dir = os.path.abspath(os.path.dirname(os.path.realpath(__file__)))
+        param_validation_path_prefix = home_dir + "/param_validation/"
+
+        param_name = type(self).__name__
+        param_validation_path = "/".join(
+            [param_validation_path_prefix, param_name + ".json"]
+        )
+
+        validation_json = None
+
+        try:
+            with open(param_validation_path, "r") as fin:
+                validation_json = json.loads(fin.read())
+        except BaseException:
+            return
+
+        self._validate_param(self, validation_json)
+
+    def _validate_param(self, param_obj, validation_json):
+        default_section = type(param_obj).__name__
+        var_list = param_obj.__dict__
+
+        for variable in var_list:
+            attr = getattr(param_obj, variable)
+
+            if type(attr).__name__ in self.builtin_types or attr is None:
+                if variable not in validation_json:
+                    continue
+
+                validation_dict = validation_json[default_section][variable]
+                value = getattr(param_obj, variable)
+                value_legal = False
+
+                for op_type in validation_dict:
+                    if self.func[op_type](value, validation_dict[op_type]):
+                        value_legal = True
+                        break
+
+                if not value_legal:
+                    raise ValueError(
+                        "Please check runtime conf, {} = {} does not match user-parameter restriction".format(
+                            variable, value
+                        )
+                    )
+
+            elif variable in validation_json:
+                self._validate_param(attr, validation_json)
+
+    @staticmethod
+    def check_string(param, descr):
+        if type(param).__name__ not in ["str"]:
+            raise ValueError(
+                descr + " {} not supported, should be string type".format(param)
+            )
+
+    @staticmethod
+    def check_empty(param, descr):
+        if not param:
+            raise ValueError(
+                descr + " does not support empty value."
+            )
+
+    @staticmethod
+    def check_positive_integer(param, descr):
+        if type(param).__name__ not in ["int", "long"] or param <= 0:
+            raise ValueError(
+                descr + " {} not supported, should be positive integer".format(param)
+            )
+
+    @staticmethod
+    def check_positive_number(param, descr):
+        if type(param).__name__ not in ["float", "int", "long"] or param <= 0:
+            raise ValueError(
+                descr + " {} not supported, should be positive numeric".format(param)
+            )
+
+    @staticmethod
+    def check_nonnegative_number(param, descr):
+        if type(param).__name__ not in ["float", "int", "long"] or param < 0:
+            raise ValueError(
+                descr
+                + " {} not supported, should be non-negative numeric".format(param)
+            )
+
+    @staticmethod
+    def check_decimal_float(param, descr):
+        if type(param).__name__ not in ["float", "int"] or param < 0 or param > 1:
+            raise ValueError(
+                descr
+                + " {} not supported, should be a float number in range [0, 1]".format(
+                    param
+                )
+            )
+
+    @staticmethod
+    def check_boolean(param, descr):
+        if type(param).__name__ != "bool":
+            raise ValueError(
+                descr + " {} not supported, should be bool type".format(param)
+            )
+
+    @staticmethod
+    def check_open_unit_interval(param, descr):
+        if type(param).__name__ not in ["float"] or param <= 0 or param >= 1:
+            raise ValueError(
+                descr + " should be a numeric number between 0 and 1 exclusively"
+            )
+
+    @staticmethod
+    def check_valid_value(param, descr, valid_values):
+        if param not in valid_values:
+            raise ValueError(
+                descr
+                + " {} is not supported, it should be in {}".format(param, valid_values)
+            )
+
+    @staticmethod
+    def check_defined_type(param, descr, types):
+        if type(param).__name__ not in types:
+            raise ValueError(
+                descr + " {} not supported, should be one of {}".format(param, types)
+            )
+
+    @staticmethod
+    def check_and_change_lower(param, valid_list, descr=""):
+        if type(param).__name__ != "str":
+            raise ValueError(
+                descr
+                + " {} not supported, should be one of {}".format(param, valid_list)
+            )
+
+        lower_param = param.lower()
+        if lower_param in valid_list:
+            return lower_param
+        else:
+            raise ValueError(
+                descr
+                + " {} not supported, should be one of {}".format(param, valid_list)
+            )
+
+    @staticmethod
+    def _greater_equal_than(value, limit):
+        return value >= limit - settings.FLOAT_ZERO
+
+    @staticmethod
+    def _less_equal_than(value, limit):
+        return value <= limit + settings.FLOAT_ZERO
+
+    @staticmethod
+    def _range(value, ranges):
+        in_range = False
+        for left_limit, right_limit in ranges:
+            if (
+                    left_limit - settings.FLOAT_ZERO
+                    <= value
+                    <= right_limit + settings.FLOAT_ZERO
+            ):
+                in_range = True
+                break
+
+        return in_range
+
+    @staticmethod
+    def _in(value, right_value_list):
+        return value in right_value_list
+
+    @staticmethod
+    def _not_in(value, wrong_value_list):
+        return value not in wrong_value_list
+
+    def _warn_deprecated_param(self, param_name, descr):
+        if self._deprecated_params_set.get(param_name):
+            logging.warning(
+                f"{descr} {param_name} is deprecated and ignored in this version."
+            )
+
+    def _warn_to_deprecate_param(self, param_name, descr, new_param):
+        if self._deprecated_params_set.get(param_name):
+            logging.warning(
+                f"{descr} {param_name} will be deprecated in future release; "
+                f"please use {new_param} instead."
+            )
+            return True
+        return False
+
+
+class ComponentBase(ABC):
+    component_name: str
+    thread_limiter = trio.CapacityLimiter(int(os.environ.get('MAX_CONCURRENT_CHATS', 10)))
+    variable_ref_patt = r"\{* *\{([a-zA-Z:0-9]+@[A-Za-z:0-9_.-]+|sys\.[a-z_]+)\} *\}*"
+
+    def __str__(self):
+        """
+        {
+            "component_name": "Begin",
+            "params": {}
+        }
+        """
+        return """{{
+            "component_name": "{}",
+            "params": {}
+        }}""".format(self.component_name,
+                     self._param
+        )
+
+    def __init__(self, canvas, id, param: ComponentParamBase):
+        from agent.canvas import Graph  # Local import to avoid cyclic dependency
+        assert isinstance(canvas, Graph), "canvas must be an instance of Canvas"
+        self._canvas = canvas
+        self._id = id
+        self._param = param
+        self._param.check()
+
+    def invoke(self, **kwargs) -> dict[str, Any]:
+        self.set_output("_created_time", time.perf_counter())
+        try:
+            self._invoke(**kwargs)
+        except Exception as e:
+            if self.get_exception_default_value():
+                self.set_exception_default_value()
+            else:
+                self.set_output("_ERROR", str(e))
+            logging.exception(e)
+        self._param.debug_inputs = {}
+        self.set_output("_elapsed_time", time.perf_counter() - self.output("_created_time"))
+        return self.output()
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60)))
+    def _invoke(self, **kwargs):
+        raise NotImplementedError()
+
+    def output(self, var_nm: str=None) -> Union[dict[str, Any], Any]:
+        if var_nm:
+            return self._param.outputs.get(var_nm, {}).get("value", "")
+        return {k: o.get("value") for k,o in self._param.outputs.items()}
+
+    def set_output(self, key: str, value: Any):
+        if key not in self._param.outputs:
+            self._param.outputs[key] = {"value": None, "type": str(type(value))}
+        self._param.outputs[key]["value"] = value
+
+    def error(self):
+        return self._param.outputs.get("_ERROR", {}).get("value")
+
+    def reset(self, only_output=False):
+        for k in self._param.outputs.keys():
+            self._param.outputs[k]["value"] = None
+        if only_output:
+            return
+        for k in self._param.inputs.keys():
+            self._param.inputs[k]["value"] = None
+        self._param.debug_inputs = {}
+
+    def get_input(self, key: str=None) -> Union[Any, dict[str, Any]]:
+        if key:
+            return self._param.inputs.get(key, {}).get("value")
+
+        res = {}
+        for var, o in self.get_input_elements().items():
+            v = self.get_param(var)
+            if v is None:
+                continue
+            if isinstance(v, str) and self._canvas.is_reff(v):
+                self.set_input_value(var, self._canvas.get_variable_value(v))
+            else:
+                self.set_input_value(var, v)
+            res[var] = self.get_input_value(var)
+        return res
+
+    def get_input_values(self) -> Union[Any, dict[str, Any]]:
+        if self._param.debug_inputs:
+            return self._param.debug_inputs
+
+        return {var: self.get_input_value(var) for var, o in self.get_input_elements().items()}
+
+    def get_input_elements_from_text(self, txt: str) -> dict[str, dict[str, str]]:
+        res = {}
+        for r in re.finditer(self.variable_ref_patt, txt, flags=re.IGNORECASE|re.DOTALL):
+            exp = r.group(1)
+            cpn_id, var_nm = exp.split("@") if exp.find("@")>0 else ("", exp)
+            res[exp] = {
+                "name": (self._canvas.get_component_name(cpn_id) +f"@{var_nm}") if cpn_id else exp,
+                "value": self._canvas.get_variable_value(exp),
+                "_retrival": self._canvas.get_variable_value(f"{cpn_id}@_references") if cpn_id else None,
+                "_cpn_id": cpn_id
+            }
+        return res
+
+    def get_input_elements(self) -> dict[str, Any]:
+        return self._param.inputs
+
+    def get_input_form(self) -> dict[str, dict]:
+        return self._param.get_input_form()
+
+    def set_input_value(self, key: str, value: Any) -> None:
+        if key not in self._param.inputs:
+            self._param.inputs[key] = {"value": None}
+        self._param.inputs[key]["value"] = value
+
+    def get_input_value(self, key: str) -> Any:
+        if key not in self._param.inputs:
+            return None
+        return self._param.inputs[key].get("value")
+
+    def get_component_name(self, cpn_id) -> str:
+        return self._canvas.get_component(cpn_id)["obj"].component_name.lower()
+
+    def get_param(self, name):
+        if hasattr(self._param, name):
+            return getattr(self._param, name)
+
+    def debug(self, **kwargs):
+        return self._invoke(**kwargs)
+
+    def get_parent(self) -> Union[object, None]:
+        pid = self._canvas.get_component(self._id).get("parent_id")
+        if not pid:
+            return
+        return self._canvas.get_component(pid)["obj"]
+
+    def get_upstream(self) -> List[str]:
+        cpn_nms = self._canvas.get_component(self._id)['upstream']
+        return cpn_nms
+
+    def get_downstream(self) -> List[str]:
+        cpn_nms = self._canvas.get_component(self._id)['downstream']
+        return cpn_nms
+
+    @staticmethod
+    def string_format(content: str, kv: dict[str, str]) -> str:
+        for n, v in kv.items():
+            def repl(_match, val=v):
+                return str(val) if val is not None else ""
+            content = re.sub(
+                r"\{%s\}" % re.escape(n),
+                repl,
+                content
+            )
+        return content
+
+    def exception_handler(self):
+        if not self._param.exception_method:
+            return
+        return {
+            "goto": self._param.exception_goto,
+            "default_value": self._param.exception_default_value
+        }
+
+    def get_exception_default_value(self):
+        if self._param.exception_method != "comment":
+            return ""
+        return self._param.exception_default_value
+
+    def set_exception_default_value(self):
+        self.set_output("result", self.get_exception_default_value())
+
+    def thoughts(self) -> str:
+        raise NotImplementedError()

agent/component/begin.py

@@ -0,0 +1,52 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+from agent.component.fillup import UserFillUpParam, UserFillUp
+
+
+class BeginParam(UserFillUpParam):
+
+    """
+    Define the Begin component parameters.
+    """
+    def __init__(self):
+        super().__init__()
+        self.mode = "conversational"
+        self.prologue = "Hi! I'm your smart assistant. What can I do for you?"
+
+    def check(self):
+        self.check_valid_value(self.mode, "The 'mode' should be either `conversational` or `task`", ["conversational", "task"])
+
+    def get_input_form(self) -> dict[str, dict]:
+        return getattr(self, "inputs")
+
+
+class Begin(UserFillUp):
+    component_name = "Begin"
+
+    def _invoke(self, **kwargs):
+        for k, v in kwargs.get("inputs", {}).items():
+            if isinstance(v, dict) and v.get("type", "").lower().find("file") >=0:
+                if v.get("optional") and v.get("value", None) is None:
+                    v = None
+                else:
+                    v = self._canvas.get_files([v["value"]])
+            else:
+                v = v.get("value")
+            self.set_output(k, v)
+            self.set_input_value(k, v)
+
+    def thoughts(self) -> str:
+        return ""

agent/component/categorize.py

@@ -0,0 +1,137 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import logging
+import os
+import re
+from abc import ABC
+
+from api.db import LLMType
+from api.db.services.llm_service import LLMBundle
+from agent.component.llm import LLMParam, LLM
+from api.utils.api_utils import timeout
+from rag.llm.chat_model import ERROR_PREFIX
+
+
+class CategorizeParam(LLMParam):
+
+    """
+    Define the categorize component parameters.
+    """
+    def __init__(self):
+        super().__init__()
+        self.category_description = {}
+        self.query = "sys.query"
+        self.message_history_window_size = 1
+        self.update_prompt()
+
+    def check(self):
+        self.check_positive_integer(self.message_history_window_size, "[Categorize] Message window size > 0")
+        self.check_empty(self.category_description, "[Categorize] Category examples")
+        for k, v in self.category_description.items():
+            if not k:
+                raise ValueError("[Categorize] Category name can not be empty!")
+            if not v.get("to"):
+                raise ValueError(f"[Categorize] 'To' of category {k} can not be empty!")
+
+    def get_input_form(self) -> dict[str, dict]:
+        return {
+            "query": {
+                "type": "line",
+                "name": "Query"
+            }
+        }
+
+    def update_prompt(self):
+        cate_lines = []
+        for c, desc in self.category_description.items():
+            for line in desc.get("examples", []):
+                if not line:
+                    continue
+                cate_lines.append("USER: \"" + re.sub(r"\n", "    ", line, flags=re.DOTALL) + "\" → "+c)
+
+        descriptions = []
+        for c, desc in self.category_description.items():
+            if desc.get("description"):
+                descriptions.append(
+                    "\n------\nCategory: {}\nDescription: {}".format(c, desc["description"]))
+
+        self.sys_prompt = """
+You are an advanced classification system that categorizes user questions into specific types. Analyze the input question and classify it into ONE of the following categories:
+{}
+
+Here's description of each category:
+ - {}
+
+---- Instructions ----
+ - Consider both explicit mentions and implied context
+ - Prioritize the most specific applicable category
+ - Return only the category name without explanations
+ - Use "Other" only when no other category fits
+
+ """.format(
+            "\n - ".join(list(self.category_description.keys())),
+            "\n".join(descriptions)
+        )
+
+        if cate_lines:
+            self.sys_prompt += """
+---- Examples ----
+{}
+""".format("\n".join(cate_lines))
+
+
+class Categorize(LLM, ABC):
+    component_name = "Categorize"
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60)))
+    def _invoke(self, **kwargs):
+        msg = self._canvas.get_history(self._param.message_history_window_size)
+        if not msg:
+            msg = [{"role": "user", "content": ""}]
+        if kwargs.get("sys.query"):
+            msg[-1]["content"] = kwargs["sys.query"]
+            self.set_input_value("sys.query", kwargs["sys.query"])
+        else:
+            msg[-1]["content"] = self._canvas.get_variable_value(self._param.query)
+            self.set_input_value(self._param.query, msg[-1]["content"])
+        self._param.update_prompt()
+        chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
+
+        user_prompt = """
+---- Real Data ----
+{} →
+""".format(" | ".join(["{}: \"{}\"".format(c["role"].upper(), re.sub(r"\n", "", c["content"], flags=re.DOTALL)) for c in msg]))
+        ans = chat_mdl.chat(self._param.sys_prompt, [{"role": "user", "content": user_prompt}], self._param.gen_conf())
+        logging.info(f"input: {user_prompt}, answer: {str(ans)}")
+        if ERROR_PREFIX in ans:
+            raise Exception(ans)
+        # Count the number of times each category appears in the answer.
+        category_counts = {}
+        for c in self._param.category_description.keys():
+            count = ans.lower().count(c.lower())
+            category_counts[c] = count
+
+        cpn_ids = list(self._param.category_description.items())[-1][1]["to"]
+        max_category = list(self._param.category_description.keys())[0]
+        if any(category_counts.values()):
+            max_category = max(category_counts.items(), key=lambda x: x[1])[0]
+            cpn_ids = self._param.category_description[max_category]["to"]
+
+        self.set_output("category_name", max_category)
+        self.set_output("_next", cpn_ids)
+
+    def thoughts(self) -> str:
+        return "Which should it falls into {}? ...".format(",".join([f"`{c}`" for c, _ in self._param.category_description.items()]))

agent/component/fillup.py

@@ -0,0 +1,40 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+from agent.component.base import ComponentBase, ComponentParamBase
+
+
+class UserFillUpParam(ComponentParamBase):
+
+    def __init__(self):
+        super().__init__()
+        self.enable_tips = True
+        self.tips = "Please fill up the form"
+
+    def check(self) -> bool:
+        return True
+
+
+class UserFillUp(ComponentBase):
+    component_name = "UserFillUp"
+
+    def _invoke(self, **kwargs):
+        for k, v in kwargs.get("inputs", {}).items():
+            self.set_output(k, v)
+
+    def thoughts(self) -> str:
+        return "Waiting for your input..."
+
+

agent/component/invoke.py

@@ -0,0 +1,135 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import json
+import logging
+import os
+import re
+import time
+from abc import ABC
+
+import requests
+
+from agent.component.base import ComponentBase, ComponentParamBase
+from api.utils.api_utils import timeout
+from deepdoc.parser import HtmlParser
+
+
+class InvokeParam(ComponentParamBase):
+    """
+    Define the Crawler component parameters.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.proxy = None
+        self.headers = ""
+        self.method = "get"
+        self.variables = []
+        self.url = ""
+        self.timeout = 60
+        self.clean_html = False
+        self.datatype = "json"  # New parameter to determine data posting type
+
+    def check(self):
+        self.check_valid_value(self.method.lower(), "Type of content from the crawler", ["get", "post", "put"])
+        self.check_empty(self.url, "End point URL")
+        self.check_positive_integer(self.timeout, "Timeout time in second")
+        self.check_boolean(self.clean_html, "Clean HTML")
+        self.check_valid_value(self.datatype.lower(), "Data post type", ["json", "formdata"])  # Check for valid datapost value
+
+
+class Invoke(ComponentBase, ABC):
+    component_name = "Invoke"
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 3)))
+    def _invoke(self, **kwargs):
+        args = {}
+        for para in self._param.variables:
+            if para.get("value"):
+                args[para["key"]] = para["value"]
+            else:
+                args[para["key"]] = self._canvas.get_variable_value(para["ref"])
+
+        url = self._param.url.strip()
+
+        def replace_variable(match):
+            var_name = match.group(1)
+            try:
+                value = self._canvas.get_variable_value(var_name)
+                return str(value or "")
+            except Exception:
+                return ""
+
+        # {base_url} or {component_id@variable_name}
+        url = re.sub(r"\{([a-zA-Z_][a-zA-Z0-9_.@-]*)\}", replace_variable, url)
+
+        if url.find("http") != 0:
+            url = "http://" + url
+
+        method = self._param.method.lower()
+        headers = {}
+        if self._param.headers:
+            headers = json.loads(self._param.headers)
+        proxies = None
+        if re.sub(r"https?:?/?/?", "", self._param.proxy):
+            proxies = {"http": self._param.proxy, "https": self._param.proxy}
+
+        last_e = ""
+        for _ in range(self._param.max_retries + 1):
+            try:
+                if method == "get":
+                    response = requests.get(url=url, params=args, headers=headers, proxies=proxies, timeout=self._param.timeout)
+                    if self._param.clean_html:
+                        sections = HtmlParser()(None, response.content)
+                        self.set_output("result", "\n".join(sections))
+                    else:
+                        self.set_output("result", response.text)
+
+                if method == "put":
+                    if self._param.datatype.lower() == "json":
+                        response = requests.put(url=url, json=args, headers=headers, proxies=proxies, timeout=self._param.timeout)
+                    else:
+                        response = requests.put(url=url, data=args, headers=headers, proxies=proxies, timeout=self._param.timeout)
+                    if self._param.clean_html:
+                        sections = HtmlParser()(None, response.content)
+                        self.set_output("result", "\n".join(sections))
+                    else:
+                        self.set_output("result", response.text)
+
+                if method == "post":
+                    if self._param.datatype.lower() == "json":
+                        response = requests.post(url=url, json=args, headers=headers, proxies=proxies, timeout=self._param.timeout)
+                    else:
+                        response = requests.post(url=url, data=args, headers=headers, proxies=proxies, timeout=self._param.timeout)
+                    if self._param.clean_html:
+                        self.set_output("result", "\n".join(sections))
+                    else:
+                        self.set_output("result", response.text)
+
+                return self.output("result")
+            except Exception as e:
+                last_e = e
+                logging.exception(f"Http request error: {e}")
+                time.sleep(self._param.delay_after_error)
+
+        if last_e:
+            self.set_output("_ERROR", str(last_e))
+            return f"Http request error: {last_e}"
+
+        assert False, self.output()
+
+    def thoughts(self) -> str:
+        return "Waiting for the server respond..."

agent/component/iteration.py

@@ -0,0 +1,60 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+from abc import ABC
+from agent.component.base import ComponentBase, ComponentParamBase
+
+
+class IterationParam(ComponentParamBase):
+    """
+    Define the Iteration component parameters.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.items_ref = ""
+
+    def get_input_form(self) -> dict[str, dict]:
+        return {
+            "items": {
+                "type": "json",
+                "name": "Items"
+            }
+        }
+
+    def check(self):
+        return True
+
+
+class Iteration(ComponentBase, ABC):
+    component_name = "Iteration"
+
+    def get_start(self):
+        for cid in self._canvas.components.keys():
+            if self._canvas.get_component(cid)["obj"].component_name.lower() != "iterationitem":
+                continue
+            if self._canvas.get_component(cid)["parent_id"] == self._id:
+                return cid
+
+    def _invoke(self, **kwargs):
+        arr = self._canvas.get_variable_value(self._param.items_ref)
+        if not isinstance(arr, list):
+            self.set_output("_ERROR", self._param.items_ref + " must be an array, but its type is "+str(type(arr)))
+
+    def thoughts(self) -> str:
+        return "Need to process {} items.".format(len(self._canvas.get_variable_value(self._param.items_ref)))
+
+
+

agent/component/iterationitem.py

@@ -0,0 +1,83 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+from abc import ABC
+from agent.component.base import ComponentBase, ComponentParamBase
+
+
+class IterationItemParam(ComponentParamBase):
+    """
+    Define the IterationItem component parameters.
+    """
+    def check(self):
+        return True
+
+
+class IterationItem(ComponentBase, ABC):
+    component_name = "IterationItem"
+
+    def __init__(self, canvas, id, param: ComponentParamBase):
+        super().__init__(canvas, id, param)
+        self._idx = 0
+
+    def _invoke(self, **kwargs):
+        parent = self.get_parent()
+        arr = self._canvas.get_variable_value(parent._param.items_ref)
+        if not isinstance(arr, list):
+            self._idx = -1
+            raise Exception(parent._param.items_ref + " must be an array, but its type is "+str(type(arr)))
+
+        if self._idx > 0:
+            self.output_collation()
+
+        if self._idx >= len(arr):
+            self._idx = -1
+            return
+
+        self.set_output("item", arr[self._idx])
+        self.set_output("index", self._idx)
+
+        self._idx += 1
+
+    def output_collation(self):
+        pid = self.get_parent()._id
+        for cid in self._canvas.components.keys():
+            obj = self._canvas.get_component_obj(cid)
+            p = obj.get_parent()
+            if not p:
+                continue
+            if p._id != pid:
+                continue
+
+            if p.component_name.lower() in ["categorize", "message", "switch", "userfillup", "interationitem"]:
+                continue
+
+            for k, o in p._param.outputs.items():
+                if "ref" not in o:
+                    continue
+                _cid, var = o["ref"].split("@")
+                if _cid != cid:
+                    continue
+                res = p.output(k)
+                if not res:
+                    res = []
+                res.append(obj.output(var))
+                p.set_output(k, res)
+
+    def end(self):
+        return self._idx == -1
+
+    def thoughts(self) -> str:
+        return "Next turn..."

agent/component/llm.py

@@ -0,0 +1,286 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import json
+import logging
+import os
+import re
+from copy import deepcopy
+from typing import Any, Generator
+import json_repair
+from functools import partial
+from api.db import LLMType
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
+from agent.component.base import ComponentBase, ComponentParamBase
+from api.utils.api_utils import timeout
+from rag.prompts.generator import tool_call_summary, message_fit_in, citation_prompt
+
+
+class LLMParam(ComponentParamBase):
+    """
+    Define the LLM component parameters.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.llm_id = ""
+        self.sys_prompt = ""
+        self.prompts = [{"role": "user", "content": "{sys.query}"}]
+        self.max_tokens = 0
+        self.temperature = 0
+        self.top_p = 0
+        self.presence_penalty = 0
+        self.frequency_penalty = 0
+        self.output_structure = None
+        self.cite = True
+        self.visual_files_var = None
+
+    def check(self):
+        self.check_decimal_float(float(self.temperature), "[Agent] Temperature")
+        self.check_decimal_float(float(self.presence_penalty), "[Agent] Presence penalty")
+        self.check_decimal_float(float(self.frequency_penalty), "[Agent] Frequency penalty")
+        self.check_nonnegative_number(int(self.max_tokens), "[Agent] Max tokens")
+        self.check_decimal_float(float(self.top_p), "[Agent] Top P")
+        self.check_empty(self.llm_id, "[Agent] LLM")
+        self.check_empty(self.sys_prompt, "[Agent] System prompt")
+        self.check_empty(self.prompts, "[Agent] User prompt")
+
+    def gen_conf(self):
+        conf = {}
+        def get_attr(nm):
+            try:
+                return getattr(self, nm)
+            except Exception:
+                pass
+
+        if int(self.max_tokens) > 0 and get_attr("maxTokensEnabled"):
+            conf["max_tokens"] = int(self.max_tokens)
+        if float(self.temperature) > 0 and get_attr("temperatureEnabled"):
+            conf["temperature"] = float(self.temperature)
+        if float(self.top_p) > 0 and get_attr("topPEnabled"):
+            conf["top_p"] = float(self.top_p)
+        if float(self.presence_penalty) > 0 and get_attr("presencePenaltyEnabled"):
+            conf["presence_penalty"] = float(self.presence_penalty)
+        if float(self.frequency_penalty) > 0 and get_attr("frequencyPenaltyEnabled"):
+            conf["frequency_penalty"] = float(self.frequency_penalty)
+        return conf
+
+
+class LLM(ComponentBase):
+    component_name = "LLM"
+
+    def __init__(self, canvas, component_id, param: ComponentParamBase):
+        super().__init__(canvas, component_id, param)
+        self.chat_mdl = LLMBundle(self._canvas.get_tenant_id(), TenantLLMService.llm_id2llm_type(self._param.llm_id),
+                                  self._param.llm_id, max_retries=self._param.max_retries,
+                                  retry_interval=self._param.delay_after_error
+                                  )
+        self.imgs = []
+
+    def get_input_form(self) -> dict[str, dict]:
+        res = {}
+        for k, v in self.get_input_elements().items():
+            res[k] = {
+                "type": "line",
+                "name": v["name"]
+            }
+        return res
+
+    def get_input_elements(self) -> dict[str, Any]:
+        res = self.get_input_elements_from_text(self._param.sys_prompt)
+        if isinstance(self._param.prompts, str):
+            self._param.prompts = [{"role": "user", "content": self._param.prompts}]
+        for prompt in self._param.prompts:
+            d = self.get_input_elements_from_text(prompt["content"])
+            res.update(d)
+        return res
+
+    def set_debug_inputs(self, inputs: dict[str, dict]):
+        self._param.debug_inputs = inputs
+
+    def add2system_prompt(self, txt):
+        self._param.sys_prompt += txt
+
+    def _sys_prompt_and_msg(self, msg, args):
+        if isinstance(self._param.prompts, str):
+            self._param.prompts = [{"role": "user", "content": self._param.prompts}]
+        for p in self._param.prompts:
+            if msg and msg[-1]["role"] == p["role"]:
+                continue
+            p = deepcopy(p)
+            p["content"] = self.string_format(p["content"], args)
+            msg.append(p)
+        return msg, self.string_format(self._param.sys_prompt, args)
+
+    def _prepare_prompt_variables(self):
+        if self._param.visual_files_var:
+            self.imgs = self._canvas.get_variable_value(self._param.visual_files_var)
+            if not self.imgs:
+                self.imgs = []
+            self.imgs = [img for img in self.imgs if img[:len("data:image/")] == "data:image/"]
+            if self.imgs and TenantLLMService.llm_id2llm_type(self._param.llm_id) == LLMType.CHAT.value:
+                self.chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT.value,
+                                          self._param.llm_id, max_retries=self._param.max_retries,
+                                          retry_interval=self._param.delay_after_error
+                                          )
+
+
+        args = {}
+        vars = self.get_input_elements() if not self._param.debug_inputs else self._param.debug_inputs
+        for k, o in vars.items():
+            args[k] = o["value"]
+            if not isinstance(args[k], str):
+                try:
+                    args[k] = json.dumps(args[k], ensure_ascii=False)
+                except Exception:
+                    args[k] = str(args[k])
+            self.set_input_value(k, args[k])
+
+        msg, sys_prompt = self._sys_prompt_and_msg(self._canvas.get_history(self._param.message_history_window_size)[:-1], args)
+        user_defined_prompt, sys_prompt = self._extract_prompts(sys_prompt)
+        if self._param.cite and self._canvas.get_reference()["chunks"]:
+            sys_prompt += citation_prompt(user_defined_prompt)
+
+        return sys_prompt, msg, user_defined_prompt
+
+    def _extract_prompts(self, sys_prompt):
+        pts = {}
+        for tag in ["TASK_ANALYSIS", "PLAN_GENERATION", "REFLECTION", "CONTEXT_SUMMARY", "CONTEXT_RANKING", "CITATION_GUIDELINES"]:
+            r = re.search(rf"<{tag}>(.*?)</{tag}>", sys_prompt, flags=re.DOTALL|re.IGNORECASE)
+            if not r:
+                continue
+            pts[tag.lower()] = r.group(1)
+            sys_prompt = re.sub(rf"<{tag}>(.*?)</{tag}>", "", sys_prompt, flags=re.DOTALL|re.IGNORECASE)
+        return pts, sys_prompt
+
+    def _generate(self, msg:list[dict], **kwargs) -> str:
+        if not self.imgs:
+            return self.chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf(), **kwargs)
+        return self.chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf(), images=self.imgs, **kwargs)
+
+    def _generate_streamly(self, msg:list[dict], **kwargs) -> Generator[str, None, None]:
+        ans = ""
+        last_idx = 0
+        endswith_think = False
+        def delta(txt):
+            nonlocal ans, last_idx, endswith_think
+            delta_ans = txt[last_idx:]
+            ans = txt
+
+            if delta_ans.find("<think>") == 0:
+                last_idx += len("<think>")
+                return "<think>"
+            elif delta_ans.find("<think>") > 0:
+                delta_ans = txt[last_idx:last_idx+delta_ans.find("<think>")]
+                last_idx += delta_ans.find("<think>")
+                return delta_ans
+            elif delta_ans.endswith("</think>"):
+                endswith_think = True
+            elif endswith_think:
+                endswith_think = False
+                return "</think>"
+
+            last_idx = len(ans)
+            if ans.endswith("</think>"):
+                last_idx -= len("</think>")
+            return re.sub(r"(<think>|</think>)", "", delta_ans)
+
+        if not self.imgs:
+            for txt in self.chat_mdl.chat_streamly(msg[0]["content"], msg[1:], self._param.gen_conf(), **kwargs):
+                yield delta(txt)
+        else:
+            for txt in self.chat_mdl.chat_streamly(msg[0]["content"], msg[1:], self._param.gen_conf(), images=self.imgs, **kwargs):
+                yield delta(txt)
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60)))
+    def _invoke(self, **kwargs):
+        def clean_formated_answer(ans: str) -> str:
+            ans = re.sub(r"^.*</think>", "", ans, flags=re.DOTALL)
+            ans = re.sub(r"^.*```json", "", ans, flags=re.DOTALL)
+            return re.sub(r"```\n*$", "", ans, flags=re.DOTALL)
+
+        prompt, msg, _ = self._prepare_prompt_variables()
+        error: str = ""
+
+        if self._param.output_structure:
+            prompt += "\nThe output MUST follow this JSON format:\n"+json.dumps(self._param.output_structure, ensure_ascii=False, indent=2)
+            prompt += "\nRedundant information is FORBIDDEN."
+            for _ in range(self._param.max_retries+1):
+                _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
+                error = ""
+                ans = self._generate(msg)
+                msg.pop(0)
+                if ans.find("**ERROR**") >= 0:
+                    logging.error(f"LLM response error: {ans}")
+                    error = ans
+                    continue
+                try:
+                    self.set_output("structured_content", json_repair.loads(clean_formated_answer(ans)))
+                    return
+                except Exception:
+                    msg.append({"role": "user", "content": "The answer can't not be parsed as JSON"})
+                    error = "The answer can't not be parsed as JSON"
+            if error:
+                self.set_output("_ERROR", error)
+            return
+
+        downstreams = self._canvas.get_component(self._id)["downstream"] if self._canvas.get_component(self._id) else []
+        ex = self.exception_handler()
+        if any([self._canvas.get_component_obj(cid).component_name.lower()=="message" for cid in downstreams]) and not self._param.output_structure and not (ex and ex["goto"]):
+            self.set_output("content", partial(self._stream_output, prompt, msg))
+            return
+
+        for _ in range(self._param.max_retries+1):
+            _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
+            error = ""
+            ans = self._generate(msg)
+            msg.pop(0)
+            if ans.find("**ERROR**") >= 0:
+                logging.error(f"LLM response error: {ans}")
+                error = ans
+                continue
+            self.set_output("content", ans)
+            break
+
+        if error:
+            if self.get_exception_default_value():
+                self.set_output("content", self.get_exception_default_value())
+            else:
+                self.set_output("_ERROR", error)
+
+    def _stream_output(self, prompt, msg):
+        _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
+        answer = ""
+        for ans in self._generate_streamly(msg):
+            if ans.find("**ERROR**") >= 0:
+                if self.get_exception_default_value():
+                    self.set_output("content", self.get_exception_default_value())
+                    yield self.get_exception_default_value()
+                else:
+                    self.set_output("_ERROR", ans)
+                return
+            yield ans
+            answer += ans
+        self.set_output("content", answer)
+
+    def add_memory(self, user:str, assist:str, func_name: str, params: dict, results: str, user_defined_prompt:dict={}):
+        summ = tool_call_summary(self.chat_mdl, func_name, params, results, user_defined_prompt)
+        logging.info(f"[MEMORY]: {summ}")
+        self._canvas.add_memory(user, assist, summ)
+
+    def thoughts(self) -> str:
+        _, msg,_ = self._prepare_prompt_variables()
+        return "⌛Give me a moment—starting from: \n\n" + re.sub(r"(User's query:|[\\]+)", '', msg[-1]['content'], flags=re.DOTALL) + "\n\nI’ll figure out our best next move."

agent/component/message.py

@@ -0,0 +1,150 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import json
+import os
+import random
+import re
+from functools import partial
+from typing import Any
+
+from agent.component.base import ComponentBase, ComponentParamBase
+from jinja2 import Template as Jinja2Template
+
+from api.utils.api_utils import timeout
+
+
+class MessageParam(ComponentParamBase):
+    """
+    Define the Message component parameters.
+    """
+    def __init__(self):
+        super().__init__()
+        self.content = []
+        self.stream = True
+        self.outputs = {
+            "content": {
+                "type": "str"
+            }
+        }
+
+    def check(self):
+        self.check_empty(self.content, "[Message] Content")
+        self.check_boolean(self.stream, "[Message] stream")
+        return True
+
+
+class Message(ComponentBase):
+    component_name = "Message"
+
+    def get_kwargs(self, script:str, kwargs:dict = {}, delimiter:str=None) -> tuple[str, dict[str, str | list | Any]]:
+        for k,v in self.get_input_elements_from_text(script).items():
+            if k in kwargs:
+                continue
+            v = v["value"]
+            if not v:
+                v = ""
+            ans = ""
+            if isinstance(v, partial):
+                for t in v():
+                    ans += t
+            elif isinstance(v, list) and delimiter:
+                ans = delimiter.join([str(vv) for vv in v])
+            elif not isinstance(v, str):
+                try:
+                    ans = json.dumps(v, ensure_ascii=False)
+                except Exception:
+                    pass
+            else:
+                ans = v
+            if not ans:
+                ans = ""
+            kwargs[k] = ans
+            self.set_input_value(k, ans)
+
+        _kwargs = {}
+        for n, v in kwargs.items():
+            _n = re.sub("[@:.]", "_", n)
+            script = re.sub(r"\{%s\}" % re.escape(n), _n, script)
+            _kwargs[_n] = v
+        return script, _kwargs
+
+    def _stream(self, rand_cnt:str):
+        s = 0
+        all_content = ""
+        cache = {}
+        for r in re.finditer(self.variable_ref_patt, rand_cnt, flags=re.DOTALL):
+            all_content += rand_cnt[s: r.start()]
+            yield rand_cnt[s: r.start()]
+            s = r.end()
+            exp = r.group(1)
+            if exp in cache:
+                yield cache[exp]
+                all_content += cache[exp]
+                continue
+
+            v = self._canvas.get_variable_value(exp)
+            if not v:
+                v = ""
+            if isinstance(v, partial):
+                cnt = ""
+                for t in v():
+                    all_content += t
+                    cnt += t
+                    yield t
+
+                continue
+            elif not isinstance(v, str):
+                try:
+                    v = json.dumps(v, ensure_ascii=False, indent=2)
+                except Exception:
+                    v = str(v)
+            yield v
+            all_content += v
+            cache[exp] = v
+
+        if s < len(rand_cnt):
+            all_content += rand_cnt[s: ]
+            yield rand_cnt[s: ]
+
+        self.set_output("content", all_content)
+
+    def _is_jinjia2(self, content:str) -> bool:
+        patt = [
+            r"\{%.*%\}", "{{", "}}"
+        ]
+        return any([re.search(p, content) for p in patt])
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60)))
+    def _invoke(self, **kwargs):
+        rand_cnt = random.choice(self._param.content)
+        if self._param.stream and not self._is_jinjia2(rand_cnt):
+            self.set_output("content", partial(self._stream, rand_cnt))
+            return
+
+        rand_cnt, kwargs = self.get_kwargs(rand_cnt, kwargs)
+        template = Jinja2Template(rand_cnt)
+        try:
+            content = template.render(kwargs)
+        except Exception:
+            pass
+
+        for n, v in kwargs.items():
+            content = re.sub(n, v, content)
+
+        self.set_output("content", content)
+
+    def thoughts(self) -> str:
+        return ""

agent/component/string_transform.py

@@ -0,0 +1,100 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import os
+import re
+from abc import ABC
+from jinja2 import Template as Jinja2Template
+from agent.component.base import ComponentParamBase
+from api.utils.api_utils import timeout
+from .message import Message
+
+
+class StringTransformParam(ComponentParamBase):
+    """
+    Define the code sandbox component parameters.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.method = "split"
+        self.script = ""
+        self.split_ref = ""
+        self.delimiters = [","]
+        self.outputs = {"result": {"value": "", "type": "string"}}
+
+    def check(self):
+        self.check_valid_value(self.method, "Support method", ["split", "merge"])
+        self.check_empty(self.delimiters, "delimiters")
+
+
+class StringTransform(Message, ABC):
+    component_name = "StringTransform"
+
+    def get_input_form(self) -> dict[str, dict]:
+        if self._param.method == "split":
+            return {
+                "line": {
+                    "name": "String",
+                    "type": "line"
+                }
+            }
+        return {k: {
+            "name": o["name"],
+            "type": "line"
+        } for k, o in self.get_input_elements_from_text(self._param.script).items()}
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60)))
+    def _invoke(self, **kwargs):
+        if self._param.method == "split":
+            self._split(kwargs.get("line"))
+        else:
+            self._merge(kwargs)
+
+    def _split(self, line:str|None = None):
+        var = self._canvas.get_variable_value(self._param.split_ref) if not line else line
+        if not var:
+            var = ""
+        assert isinstance(var, str), "The input variable is not a string: {}".format(type(var))
+        self.set_input_value(self._param.split_ref, var)
+        res = []
+        for i,s in enumerate(re.split(r"(%s)"%("|".join([re.escape(d) for d in self._param.delimiters])), var, flags=re.DOTALL)):
+            if i % 2 == 1:
+                continue
+            res.append(s)
+        self.set_output("result", res)
+
+    def _merge(self, kwargs:dict[str, str] = {}):
+        script = self._param.script
+        script, kwargs = self.get_kwargs(script, kwargs, self._param.delimiters[0])
+
+        if self._is_jinjia2(script):
+            template = Jinja2Template(script)
+            try:
+                script = template.render(kwargs)
+            except Exception:
+                pass
+
+        for k,v in kwargs.items():
+            if not v:
+                v = ""
+            script = re.sub(k, lambda match: v, script)
+
+        self.set_output("result", script)
+
+    def thoughts(self) -> str:
+        return f"It's {self._param.method}ing."
+
+

agent/component/switch.py

@@ -0,0 +1,131 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import numbers
+import os
+from abc import ABC
+from typing import Any
+
+from agent.component.base import ComponentBase, ComponentParamBase
+from api.utils.api_utils import timeout
+
+
+class SwitchParam(ComponentParamBase):
+    """
+    Define the Switch component parameters.
+    """
+
+    def __init__(self):
+        super().__init__()
+        """
+        {
+            "logical_operator" : "and | or"
+            "items" : [
+                            {"cpn_id": "categorize:0", "operator": "contains", "value": ""},
+                            {"cpn_id": "categorize:0", "operator": "contains", "value": ""},...],
+            "to": ""
+        }
+        """
+        self.conditions = []
+        self.end_cpn_ids = []
+        self.operators = ['contains', 'not contains', 'start with', 'end with', 'empty', 'not empty', '=', '≠', '>',
+                          '<', '≥', '≤']
+
+    def check(self):
+        self.check_empty(self.conditions, "[Switch] conditions")
+        for cond in self.conditions:
+            if not cond["to"]:
+                raise ValueError("[Switch] 'To' can not be empty!")
+        self.check_empty(self.end_cpn_ids, "[Switch] the ELSE/Other destination can not be empty.")
+
+    def get_input_form(self) -> dict[str, dict]:
+        return {
+            "urls": {
+                "name": "URLs",
+                "type": "line"
+            }
+        }
+
+class Switch(ComponentBase, ABC):
+    component_name = "Switch"
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 3)))
+    def _invoke(self, **kwargs):
+        for cond in self._param.conditions:
+            res = []
+            for item in cond["items"]:
+                if not item["cpn_id"]:
+                    continue
+                cpn_v = self._canvas.get_variable_value(item["cpn_id"])
+                self.set_input_value(item["cpn_id"], cpn_v)
+                operatee = item.get("value", "")
+                if isinstance(cpn_v, numbers.Number):
+                    operatee = float(operatee)
+                res.append(self.process_operator(cpn_v, item["operator"], operatee))
+                if cond["logical_operator"] != "and" and any(res):
+                    self.set_output("next", [self._canvas.get_component_name(cpn_id) for cpn_id in cond["to"]])
+                    self.set_output("_next", cond["to"])
+                    return
+
+            if all(res):
+                self.set_output("next", [self._canvas.get_component_name(cpn_id) for cpn_id in cond["to"]])
+                self.set_output("_next", cond["to"])
+                return
+
+        self.set_output("next", [self._canvas.get_component_name(cpn_id) for cpn_id in self._param.end_cpn_ids])
+        self.set_output("_next", self._param.end_cpn_ids)
+
+    def process_operator(self, input: Any, operator: str, value: Any) -> bool:
+        if operator == "contains":
+            return True if value.lower() in input.lower() else False
+        elif operator == "not contains":
+            return True if value.lower() not in input.lower() else False
+        elif operator == "start with":
+            return True if input.lower().startswith(value.lower()) else False
+        elif operator == "end with":
+            return True if input.lower().endswith(value.lower()) else False
+        elif operator == "empty":
+            return True if not input else False
+        elif operator == "not empty":
+            return True if input else False
+        elif operator == "=":
+            return True if input == value else False
+        elif operator == "≠":
+            return True if input != value else False
+        elif operator == ">":
+            try:
+                return True if float(input) > float(value) else False
+            except Exception:
+                return True if input > value else False
+        elif operator == "<":
+            try:
+                return True if float(input) < float(value) else False
+            except Exception:
+                return True if input < value else False
+        elif operator == "≥":
+            try:
+                return True if float(input) >= float(value) else False
+            except Exception:
+                return True if input >= value else False
+        elif operator == "≤":
+            try:
+                return True if float(input) <= float(value) else False
+            except Exception:
+                return True if input <= value else False
+
+        raise ValueError('Not supported operator' + operator)
+
+    def thoughts(self) -> str:
+        return "I’m weighing a few options and will pick the next step shortly."