lfx-nightly 0.2.0.dev25__py3-none-any.whl

lfx/components/cassandra/cassandra_graph.py

@@ -0,0 +1,238 @@
+from uuid import UUID
+
+from langchain_community.graph_vectorstores import CassandraGraphVectorStore
+
+from lfx.base.vectorstores.model import LCVectorStoreComponent, check_cached_vector_store
+from lfx.helpers.data import docs_to_data
+from lfx.inputs.inputs import DictInput, FloatInput
+from lfx.io import (
+    DropdownInput,
+    HandleInput,
+    IntInput,
+    MessageTextInput,
+    SecretStrInput,
+)
+from lfx.schema.data import Data
+
+
+class CassandraGraphVectorStoreComponent(LCVectorStoreComponent):
+    display_name = "Cassandra Graph"
+    description = "Cassandra Graph Vector Store"
+    name = "CassandraGraph"
+    icon = "Cassandra"
+
+    inputs = [
+        MessageTextInput(
+            name="database_ref",
+            display_name="Contact Points / Astra Database ID",
+            info="Contact points for the database (or Astra DB database ID)",
+            required=True,
+        ),
+        MessageTextInput(
+            name="username", display_name="Username", info="Username for the database (leave empty for Astra DB)."
+        ),
+        SecretStrInput(
+            name="token",
+            display_name="Password / Astra DB Token",
+            info="User password for the database (or Astra DB token).",
+            required=True,
+        ),
+        MessageTextInput(
+            name="keyspace",
+            display_name="Keyspace",
+            info="Table Keyspace (or Astra DB namespace).",
+            required=True,
+        ),
+        MessageTextInput(
+            name="table_name",
+            display_name="Table Name",
+            info="The name of the table (or Astra DB collection) where vectors will be stored.",
+            required=True,
+        ),
+        DropdownInput(
+            name="setup_mode",
+            display_name="Setup Mode",
+            info="Configuration mode for setting up the Cassandra table, with options like 'Sync' or 'Off'.",
+            options=["Sync", "Off"],
+            value="Sync",
+            advanced=True,
+        ),
+        DictInput(
+            name="cluster_kwargs",
+            display_name="Cluster arguments",
+            info="Optional dictionary of additional keyword arguments for the Cassandra cluster.",
+            advanced=True,
+            list=True,
+        ),
+        *LCVectorStoreComponent.inputs,
+        HandleInput(name="embedding", display_name="Embedding", input_types=["Embeddings"]),
+        IntInput(
+            name="number_of_results",
+            display_name="Number of Results",
+            info="Number of results to return.",
+            value=4,
+            advanced=True,
+        ),
+        DropdownInput(
+            name="search_type",
+            display_name="Search Type",
+            info="Search type to use",
+            options=[
+                "Traversal",
+                "MMR traversal",
+                "Similarity",
+                "Similarity with score threshold",
+                "MMR (Max Marginal Relevance)",
+            ],
+            value="Traversal",
+            advanced=True,
+        ),
+        IntInput(
+            name="depth",
+            display_name="Depth of traversal",
+            info="The maximum depth of edges to traverse. (when using 'Traversal' or 'MMR traversal')",
+            value=1,
+            advanced=True,
+        ),
+        FloatInput(
+            name="search_score_threshold",
+            display_name="Search Score Threshold",
+            info="Minimum similarity score threshold for search results. "
+            "(when using 'Similarity with score threshold')",
+            value=0,
+            advanced=True,
+        ),
+        DictInput(
+            name="search_filter",
+            display_name="Search Metadata Filter",
+            info="Optional dictionary of filters to apply to the search query.",
+            advanced=True,
+            list=True,
+        ),
+    ]
+
+    @check_cached_vector_store
+    def build_vector_store(self) -> CassandraGraphVectorStore:
+        try:
+            import cassio
+            from langchain_community.utilities.cassandra import SetupMode
+        except ImportError as e:
+            msg = "Could not import cassio integration package. Please install it with `pip install cassio`."
+            raise ImportError(msg) from e
+
+        database_ref = self.database_ref
+
+        try:
+            UUID(self.database_ref)
+            is_astra = True
+        except ValueError:
+            is_astra = False
+            if "," in self.database_ref:
+                # use a copy because we can't change the type of the parameter
+                database_ref = self.database_ref.split(",")
+
+        if is_astra:
+            cassio.init(
+                database_id=database_ref,
+                token=self.token,
+                cluster_kwargs=self.cluster_kwargs,
+            )
+        else:
+            cassio.init(
+                contact_points=database_ref,
+                username=self.username,
+                password=self.token,
+                cluster_kwargs=self.cluster_kwargs,
+            )
+
+        self.ingest_data = self._prepare_ingest_data()
+
+        documents = []
+
+        for _input in self.ingest_data or []:
+            if isinstance(_input, Data):
+                documents.append(_input.to_lc_document())
+            else:
+                documents.append(_input)
+
+        setup_mode = SetupMode.OFF if self.setup_mode == "Off" else SetupMode.SYNC
+
+        if documents:
+            self.log(f"Adding {len(documents)} documents to the Vector Store.")
+            store = CassandraGraphVectorStore.from_documents(
+                documents=documents,
+                embedding=self.embedding,
+                node_table=self.table_name,
+                keyspace=self.keyspace,
+            )
+        else:
+            self.log("No documents to add to the Vector Store.")
+            store = CassandraGraphVectorStore(
+                embedding=self.embedding,
+                node_table=self.table_name,
+                keyspace=self.keyspace,
+                setup_mode=setup_mode,
+            )
+        return store
+
+    def _map_search_type(self) -> str:
+        if self.search_type == "Similarity":
+            return "similarity"
+        if self.search_type == "Similarity with score threshold":
+            return "similarity_score_threshold"
+        if self.search_type == "MMR (Max Marginal Relevance)":
+            return "mmr"
+        if self.search_type == "MMR Traversal":
+            return "mmr_traversal"
+        return "traversal"
+
+    def search_documents(self) -> list[Data]:
+        vector_store = self.build_vector_store()
+
+        self.log(f"Search input: {self.search_query}")
+        self.log(f"Search type: {self.search_type}")
+        self.log(f"Number of results: {self.number_of_results}")
+
+        if self.search_query and isinstance(self.search_query, str) and self.search_query.strip():
+            try:
+                search_type = self._map_search_type()
+                search_args = self._build_search_args()
+
+                self.log(f"Search args: {search_args}")
+
+                docs = vector_store.search(query=self.search_query, search_type=search_type, **search_args)
+            except KeyError as e:
+                if "content" in str(e):
+                    msg = (
+                        "You should ingest data through Langflow (or LangChain) to query it in Langflow. "
+                        "Your collection does not contain a field name 'content'."
+                    )
+                    raise ValueError(msg) from e
+                raise
+
+            self.log(f"Retrieved documents: {len(docs)}")
+
+            data = docs_to_data(docs)
+            self.status = data
+            return data
+        return []
+
+    def _build_search_args(self):
+        args = {
+            "k": self.number_of_results,
+            "score_threshold": self.search_score_threshold,
+            "depth": self.depth,
+        }
+
+        if self.search_filter:
+            clean_filter = {k: v for k, v in self.search_filter.items() if k and v}
+            if len(clean_filter) > 0:
+                args["filter"] = clean_filter
+        return args
+
+    def get_retriever_kwargs(self):
+        search_args = self._build_search_args()
+        return {
+            "search_type": self._map_search_type(),
+            "search_kwargs": search_args,
+        }

lfx/components/chains/__init__.py

@@ -0,0 +1,3 @@
+"""LangFlow chains components."""
+
+__all__: list[str] = []

lfx/components/chroma/__init__.py

@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+from lfx.components._importing import import_mod
+
+if TYPE_CHECKING:
+    from .chroma import ChromaVectorStoreComponent
+
+_dynamic_imports = {
+    "ChromaVectorStoreComponent": "chroma",
+}
+
+__all__ = [
+    "ChromaVectorStoreComponent",
+]
+
+
+def __getattr__(attr_name: str) -> Any:
+    """Lazily import Chroma components on attribute access."""
+    if attr_name not in _dynamic_imports:
+        msg = f"module '{__name__}' has no attribute '{attr_name}'"
+        raise AttributeError(msg)
+    try:
+        result = import_mod(attr_name, _dynamic_imports[attr_name], __spec__.parent)
+    except (ModuleNotFoundError, ImportError, AttributeError) as e:
+        msg = f"Could not import '{attr_name}' from '{__name__}': {e}"
+        raise AttributeError(msg) from e
+    globals()[attr_name] = result
+    return result
+
+
+def __dir__() -> list[str]:
+    return list(__all__)

lfx/components/chroma/chroma.py

@@ -0,0 +1,169 @@
+from copy import deepcopy
+from typing import TYPE_CHECKING
+
+from chromadb.config import Settings
+from langchain_chroma import Chroma
+from typing_extensions import override
+
+from lfx.base.vectorstores.model import LCVectorStoreComponent, check_cached_vector_store
+from lfx.base.vectorstores.utils import chroma_collection_to_data
+from lfx.inputs.inputs import BoolInput, DropdownInput, HandleInput, IntInput, StrInput
+from lfx.schema.data import Data
+
+if TYPE_CHECKING:
+    from lfx.schema.dataframe import DataFrame
+
+
+class ChromaVectorStoreComponent(LCVectorStoreComponent):
+    """Chroma Vector Store with search capabilities."""
+
+    display_name: str = "Chroma DB"
+    description: str = "Chroma Vector Store with search capabilities"
+    name = "Chroma"
+    icon = "Chroma"
+
+    inputs = [
+        StrInput(
+            name="collection_name",
+            display_name="Collection Name",
+            value="langflow",
+        ),
+        StrInput(
+            name="persist_directory",
+            display_name="Persist Directory",
+        ),
+        *LCVectorStoreComponent.inputs,
+        HandleInput(name="embedding", display_name="Embedding", input_types=["Embeddings"]),
+        StrInput(
+            name="chroma_server_cors_allow_origins",
+            display_name="Server CORS Allow Origins",
+            advanced=True,
+        ),
+        StrInput(
+            name="chroma_server_host",
+            display_name="Server Host",
+            advanced=True,
+        ),
+        IntInput(
+            name="chroma_server_http_port",
+            display_name="Server HTTP Port",
+            advanced=True,
+        ),
+        IntInput(
+            name="chroma_server_grpc_port",
+            display_name="Server gRPC Port",
+            advanced=True,
+        ),
+        BoolInput(
+            name="chroma_server_ssl_enabled",
+            display_name="Server SSL Enabled",
+            advanced=True,
+        ),
+        BoolInput(
+            name="allow_duplicates",
+            display_name="Allow Duplicates",
+            advanced=True,
+            info="If false, will not add documents that are already in the Vector Store.",
+        ),
+        DropdownInput(
+            name="search_type",
+            display_name="Search Type",
+            options=["Similarity", "MMR"],
+            value="Similarity",
+            advanced=True,
+        ),
+        IntInput(
+            name="number_of_results",
+            display_name="Number of Results",
+            info="Number of results to return.",
+            advanced=True,
+            value=10,
+        ),
+        IntInput(
+            name="limit",
+            display_name="Limit",
+            advanced=True,
+            info="Limit the number of records to compare when Allow Duplicates is False.",
+        ),
+    ]
+
+    @override
+    @check_cached_vector_store
+    def build_vector_store(self) -> Chroma:
+        """Builds the Chroma object."""
+        try:
+            from chromadb import Client
+            from langchain_chroma import Chroma
+        except ImportError as e:
+            msg = "Could not import Chroma integration package. Please install it with `pip install langchain-chroma`."
+            raise ImportError(msg) from e
+        # Chroma settings
+        chroma_settings = None
+        client = None
+        if self.chroma_server_host:
+            chroma_settings = Settings(
+                chroma_server_cors_allow_origins=self.chroma_server_cors_allow_origins or [],
+                chroma_server_host=self.chroma_server_host,
+                chroma_server_http_port=self.chroma_server_http_port or None,
+                chroma_server_grpc_port=self.chroma_server_grpc_port or None,
+                chroma_server_ssl_enabled=self.chroma_server_ssl_enabled,
+            )
+            client = Client(settings=chroma_settings)
+
+        # Check persist_directory and expand it if it is a relative path
+        persist_directory = self.resolve_path(self.persist_directory) if self.persist_directory is not None else None
+
+        chroma = Chroma(
+            persist_directory=persist_directory,
+            client=client,
+            embedding_function=self.embedding,
+            collection_name=self.collection_name,
+        )
+
+        self._add_documents_to_vector_store(chroma)
+        limit = int(self.limit) if self.limit is not None and str(self.limit).strip() else None
+        self.status = chroma_collection_to_data(chroma.get(limit=limit))
+        return chroma
+
+    def _add_documents_to_vector_store(self, vector_store: "Chroma") -> None:
+        """Adds documents to the Vector Store."""
+        ingest_data: list | Data | DataFrame = self.ingest_data
+        if not ingest_data:
+            self.status = ""
+            return
+
+        # Convert DataFrame to Data if needed using parent's method
+        ingest_data = self._prepare_ingest_data()
+
+        stored_documents_without_id = []
+        if self.allow_duplicates:
+            stored_data = []
+        else:
+            limit = int(self.limit) if self.limit is not None and str(self.limit).strip() else None
+            stored_data = chroma_collection_to_data(vector_store.get(limit=limit))
+            for value in deepcopy(stored_data):
+                del value.id
+                stored_documents_without_id.append(value)
+
+        documents = []
+        for _input in ingest_data or []:
+            if isinstance(_input, Data):
+                if _input not in stored_documents_without_id:
+                    documents.append(_input.to_lc_document())
+            else:
+                msg = "Vector Store Inputs must be Data objects."
+                raise TypeError(msg)
+
+        if documents and self.embedding is not None:
+            self.log(f"Adding {len(documents)} documents to the Vector Store.")
+            # Filter complex metadata to prevent ChromaDB errors
+            try:
+                from langchain_community.vectorstores.utils import filter_complex_metadata
+
+                filtered_documents = filter_complex_metadata(documents)
+                vector_store.add_documents(filtered_documents)
+            except ImportError:
+                self.log("Warning: Could not import filter_complex_metadata. Adding documents without filtering.")
+                vector_store.add_documents(documents)
+        else:
+            self.log("No documents to add to the Vector Store.")

lfx/components/cleanlab/__init__.py

@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+from lfx.components._importing import import_mod
+
+if TYPE_CHECKING:
+    from .cleanlab_evaluator import CleanlabEvaluator
+    from .cleanlab_rag_evaluator import CleanlabRAGEvaluator
+    from .cleanlab_remediator import CleanlabRemediator
+
+_dynamic_imports = {
+    "CleanlabEvaluator": "cleanlab_evaluator",
+    "CleanlabRAGEvaluator": "cleanlab_rag_evaluator",
+    "CleanlabRemediator": "cleanlab_remediator",
+}
+
+__all__ = [
+    "CleanlabEvaluator",
+    "CleanlabRAGEvaluator",
+    "CleanlabRemediator",
+]
+
+
+def __getattr__(attr_name: str) -> Any:
+    """Lazily import cleanlab components on attribute access."""
+    if attr_name not in _dynamic_imports:
+        msg = f"module '{__name__}' has no attribute '{attr_name}'"
+        raise AttributeError(msg)
+    try:
+        result = import_mod(attr_name, _dynamic_imports[attr_name], __spec__.parent)
+    except (ModuleNotFoundError, ImportError, AttributeError) as e:
+        msg = f"Could not import '{attr_name}' from '{__name__}': {e}"
+        raise AttributeError(msg) from e
+    globals()[attr_name] = result
+    return result
+
+
+def __dir__() -> list[str]:
+    return list(__all__)

lfx/components/cleanlab/cleanlab_evaluator.py

@@ -0,0 +1,155 @@
+from cleanlab_tlm import TLM
+
+from lfx.custom import Component
+from lfx.io import (
+    DropdownInput,
+    MessageTextInput,
+    Output,
+    SecretStrInput,
+)
+from lfx.schema.message import Message
+
+
+class CleanlabEvaluator(Component):
+    """A component that evaluates the trustworthiness of LLM responses using Cleanlab.
+
+    This component takes a prompt and response pair, along with optional system instructions,
+    and uses Cleanlab's evaluation algorithms to generate a trust score and explanation.
+
+    Inputs:
+        - system_prompt (MessageTextInput): Optional system-level instructions prepended to the user prompt.
+        - prompt (MessageTextInput): The user's prompt or query sent to the LLM.
+        - response (MessageTextInput): The response generated by the LLM to be evaluated. This should come from the
+          LLM component, i.e. OpenAI, Gemini, etc.
+        - api_key (SecretStrInput): Your Cleanlab API key.
+        - model (DropdownInput): The model used by Cleanlab to evaluate the response (can differ from the
+          generation model).
+        - quality_preset (DropdownInput): Tradeoff setting for accuracy vs. speed and cost. Higher presets are
+          slower but more accurate.
+
+    Outputs:
+        - response_passthrough (Message): The original response, passed through for downstream use.
+        - score (number): A float between 0 and 1 indicating Cleanlab's trustworthiness score for the response.
+        - explanation (Message): A textual explanation of why the response received its score.
+
+    This component works well in conjunction with the CleanlabRemediator to create a complete trust evaluation
+    and remediation pipeline.
+
+    More details on the evaluation metrics can be found here: https://help.cleanlab.ai/tlm/tutorials/tlm/
+    """
+
+    display_name = "Cleanlab Evaluator"
+    description = "Evaluates any LLM response using Cleanlab and outputs trust score and explanation."
+    icon = "Cleanlab"
+    name = "CleanlabEvaluator"
+
+    inputs = [
+        MessageTextInput(
+            name="system_prompt",
+            display_name="System Message",
+            info="System-level instructions prepended to the user query.",
+            value="",
+        ),
+        MessageTextInput(
+            name="prompt",
+            display_name="Prompt",
+            info="The user's query to the model.",
+            required=True,
+        ),
+        MessageTextInput(
+            name="response",
+            display_name="Response",
+            info="The response to the user's query.",
+            required=True,
+        ),
+        SecretStrInput(
+            name="api_key",
+            display_name="Cleanlab API Key",
+            info="Your Cleanlab API key.",
+            required=True,
+        ),
+        DropdownInput(
+            name="model",
+            display_name="Cleanlab Evaluation Model",
+            options=[
+                "gpt-4.1",
+                "gpt-4.1-mini",
+                "gpt-4.1-nano",
+                "o4-mini",
+                "o3",
+                "gpt-4.5-preview",
+                "gpt-4o-mini",
+                "gpt-4o",
+                "o3-mini",
+                "o1",
+                "o1-mini",
+                "gpt-4",
+                "gpt-3.5-turbo-16k",
+                "claude-3.7-sonnet",
+                "claude-3.5-sonnet-v2",
+                "claude-3.5-sonnet",
+                "claude-3.5-haiku",
+                "claude-3-haiku",
+                "nova-micro",
+                "nova-lite",
+                "nova-pro",
+            ],
+            info="The model Cleanlab uses to evaluate the response. This does NOT need to be the same model that "
+            "generated the response.",
+            value="gpt-4o-mini",
+            required=True,
+            advanced=True,
+        ),
+        DropdownInput(
+            name="quality_preset",
+            display_name="Quality Preset",
+            options=["base", "low", "medium", "high", "best"],
+            value="medium",
+            info="This determines the accuracy, latency, and cost of the evaluation. Higher quality is generally "
+            "slower but more accurate.",
+            required=True,
+            advanced=True,
+        ),
+    ]
+
+    outputs = [
+        Output(
+            display_name="Response",
+            name="response_passthrough",
+            method="pass_response",
+            types=["Message"],
+        ),
+        Output(display_name="Trust Score", name="score", method="get_score", types=["number"]),
+        Output(
+            display_name="Explanation",
+            name="explanation",
+            method="get_explanation",
+            types=["Message"],
+        ),
+    ]
+
+    def _evaluate_once(self):
+        if not hasattr(self, "_cached_result"):
+            full_prompt = f"{self.system_prompt}\n\n{self.prompt}" if self.system_prompt else self.prompt
+            tlm = TLM(
+                api_key=self.api_key,
+                options={"log": ["explanation"], "model": self.model},
+                quality_preset=self.quality_preset,
+            )
+            self._cached_result = tlm.get_trustworthiness_score(full_prompt, self.response)
+        return self._cached_result
+
+    def get_score(self) -> float:
+        result = self._evaluate_once()
+        score = result.get("trustworthiness_score", 0.0)
+        self.status = f"Trust score: {score:.2f}"
+        return score
+
+    def get_explanation(self) -> Message:
+        result = self._evaluate_once()
+        explanation = result.get("log", {}).get("explanation", "No explanation returned.")
+        return Message(text=explanation)
+
+    def pass_response(self) -> Message:
+        self.status = "Passing through response."
+        return Message(text=self.response)