Re-order extraction metadata union for better parsing (#865)

* Re-order args so that pydantic doesn't parse nested dict to a empty extraction result

* Use a citations array instead

py/llama_cloud_services/beta/agent_data/schema.py

@@ -37,11 +37,10 @@ Example Usage:
 """
 
 from datetime import datetime
-import numbers
 from llama_cloud import ExtractRun
 from llama_cloud.types.agent_data import AgentData
 from llama_cloud.types.aggregate_group import AggregateGroup
-from pydantic import BaseModel, Field, ValidationError
+from pydantic import BaseModel, Field, ValidationError, model_validator, ConfigDict
 from typing import (
     Generic,
     List,
@@ -176,6 +175,16 @@ class TypedAgentDataItems(BaseModel, Generic[AgentDataT]):
     )
 
 
+class FieldCitation(BaseModel):
+    page: Optional[int] = Field(
+        None, description="The page number that the field occurred on"
+    )
+    matching_text: Optional[str] = Field(
+        None,
+        description="The original text this field's value was derived from",
+    )
+
+
 class ExtractedFieldMetadata(BaseModel):
     """
     Metadata for an extracted data field, such as confidence, and citation information.
@@ -193,14 +202,14 @@ class ExtractedFieldMetadata(BaseModel):
         None,
         description="The confidence score for the field based on the extracted text only",
     )
-    page_number: Optional[int] = Field(
-        None, description="The page number that the field occurred on"
-    )
-    matching_text: Optional[str] = Field(
+    citation: Optional[List[FieldCitation]] = Field(
         None,
-        description="The original text this field's value was derived from",
+        description="The citation for the field, including page number and matching text",
     )
 
+    # Forbid unknown keys to avoid swallowing nested dicts
+    model_config = ConfigDict(extra="forbid")
+
 
 ExtractedFieldMetaDataDict = Dict[
     str, Union[ExtractedFieldMetadata, Dict[str, Any], list[Any]]
@@ -238,19 +247,10 @@ def _parse_extracted_field_metadata_recursive(
         if len(indicator_fields.intersection(field_value.keys())) > 0:
             try:
                 merged = {**field_value, **additional_fields}
+                allowed_fields = ExtractedFieldMetadata.model_fields.keys()
+                merged = {k: v for k, v in merged.items() if k in allowed_fields}
                 validated = ExtractedFieldMetadata.model_validate(merged)
 
-                # grab the citation from the array. This is just an array for backwards compatibility.
-                if "citation" in field_value and len(field_value["citation"]) > 0:
-                    first_citation = field_value["citation"][0]
-                    if "page" in first_citation and isinstance(
-                        first_citation["page"], numbers.Number
-                    ):
-                        validated.page_number = int(first_citation["page"])  # type: ignore
-                    if "matching_text" in first_citation and isinstance(
-                        first_citation["matching_text"], str
-                    ):
-                        validated.matching_text = first_citation["matching_text"]
                 return validated
             except ValidationError:
                 pass
@@ -340,6 +340,28 @@ class ExtractedData(BaseModel, Generic[ExtractedT]):
         description="Additional metadata about the extracted data, such as errors, tokens, etc.",
     )
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_field_metadata_on_input(cls, value: Any) -> Any:
+        # Ensure any inbound representation (including JSON round-trips)
+        # gets normalized so nested dicts become ExtractedFieldMetadata where appropriate.
+        if (
+            isinstance(value, dict)
+            and "field_metadata" in value
+            and isinstance(value["field_metadata"], dict)
+        ):
+            try:
+                value = {
+                    **value,
+                    "field_metadata": parse_extracted_field_metadata(
+                        value["field_metadata"]
+                    ),
+                }
+            except Exception:
+                # Let pydantic surface detailed errors later rather than swallowing completely
+                pass
+        return value
+
     @classmethod
     def create(
         cls,

py/llama_cloud_services/beta/classifier/client.py

@@ -0,0 +1,187 @@
+import asyncio
+import time
+from typing import Optional
+from pydantic import BaseModel
+from llama_cloud.client import AsyncLlamaCloud
+from llama_cloud.types import (
+    ClassifierRule,
+    ClassifyJobResults,
+    ClassifyParsingConfiguration,
+    StatusEnum,
+    ClassifyJobWithStatus,
+    File,
+)
+from llama_cloud.resources.classifier.client import OMIT
+from llama_cloud_services.files.client import FileClient
+from llama_cloud_services.constants import POLLING_TIMEOUT_SECONDS
+from llama_cloud_services.utils import is_terminal_status, augment_async_errors
+from llama_index.core.async_utils import DEFAULT_NUM_WORKERS, run_jobs
+
+
+class ClassificationOutput(BaseModel):
+    file_id: str
+    classification: str
+
+
+class ClassifyClient:
+    """
+    Experimental - Client for interacting with the LlamaCloud Classifier API.
+    The Classification API is currently in beta and may change in the future without notice.
+
+    Args:
+        client: The LlamaCloud client to use.
+        project_id: The project ID to use.
+        organization_id: The organization ID to use.
+        polling_interval: The interval to poll for job completion in seconds.
+        polling_timeout: The timeout for the job to complete in seconds.
+    """
+
+    def __init__(
+        self,
+        client: AsyncLlamaCloud,
+        project_id: Optional[str] = None,
+        organization_id: Optional[str] = None,
+        polling_interval: float = 1.0,
+        polling_timeout: float = POLLING_TIMEOUT_SECONDS,
+    ):
+        self.client = client
+        self.project_id = project_id
+        self.organization_id = organization_id
+        self.polling_interval = polling_interval
+        self.file_client = FileClient(client, project_id, organization_id)
+        self.polling_timeout = polling_timeout
+
+    async def aclassify_file_ids(
+        self,
+        rules: list[ClassifierRule],
+        file_ids: list[str],
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+    ) -> ClassifyJobResults:
+        """
+        Classify a list of files by their IDs.
+        Note that even if a job fails, some of the files may have been classified successfully.
+        In this case, you may want to set raise_on_error to False and check the results for successful classifications.
+
+        Args:
+            rules: The rules to use for classification.
+            file_ids: The IDs of the files to classify.
+            parsing_configuration: The parsing configuration to use for classification.
+            raise_on_error: Whether to raise an error if the classification job fails.
+
+        Returns:
+            The results of the classification job.
+        """
+        classify_job = await self.client.classifier.create_classify_job(
+            rules=rules,
+            file_ids=file_ids,
+            parsing_configuration=parsing_configuration or OMIT,
+            project_id=self.project_id,
+            organization_id=self.organization_id,
+        )
+
+        classify_job_with_status = await self._wait_for_job_completion(classify_job.id)
+
+        if raise_on_error and classify_job_with_status.status == StatusEnum.ERROR:
+            raise ValueError(
+                f"Error classifying files under job ID {classify_job_with_status.id}"
+            )
+
+        results = await self.client.classifier.get_classification_job_results(
+            classify_job_with_status.id,
+            project_id=self.project_id,
+            organization_id=self.organization_id,
+        )
+
+        return results
+
+    def classify_file_ids(
+        self,
+        rules: list[ClassifierRule],
+        file_ids: list[str],
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+    ) -> ClassifyJobResults:
+        with augment_async_errors():
+            return asyncio.run(
+                self.aclassify_file_ids(
+                    rules, file_ids, parsing_configuration, raise_on_error
+                )
+            )
+
+    async def aclassify_file_path(
+        self,
+        rules: list[ClassifierRule],
+        file_input_path: str,
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+    ) -> ClassifyJobResults:
+        file = await self.file_client.upload_file(file_input_path)
+        return await self.aclassify_file_ids(
+            rules, [file.id], parsing_configuration, raise_on_error
+        )
+
+    def classify_file_path(
+        self,
+        rules: list[ClassifierRule],
+        file_input_path: str,
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+    ) -> ClassifyJobResults:
+        with augment_async_errors():
+            return asyncio.run(
+                self.aclassify_file_path(
+                    rules, file_input_path, parsing_configuration, raise_on_error
+                )
+            )
+
+    async def aclassify_file_paths(
+        self,
+        rules: list[ClassifierRule],
+        file_input_paths: list[str],
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+        workers: int = DEFAULT_NUM_WORKERS,
+        show_progress: bool = False,
+    ) -> ClassifyJobResults:
+        coroutines = [self.file_client.upload_file(path) for path in file_input_paths]
+        files: list[File] = await run_jobs(
+            coroutines,
+            show_progress=show_progress,
+            workers=workers,
+            desc="Uploading files for classification",
+        )
+        return await self.aclassify_file_ids(
+            rules, [file.id for file in files], parsing_configuration, raise_on_error
+        )
+
+    def classify_file_paths(
+        self,
+        rules: list[ClassifierRule],
+        file_input_paths: list[str],
+        parsing_configuration: Optional[ClassifyParsingConfiguration] = None,
+        raise_on_error: bool = True,
+    ) -> ClassifyJobResults:
+        with augment_async_errors():
+            return asyncio.run(
+                self.aclassify_file_paths(
+                    rules, file_input_paths, parsing_configuration, raise_on_error
+                )
+            )
+
+    async def _wait_for_job_completion(self, job_id: str) -> ClassifyJobWithStatus:
+        job = await self.client.classifier.get_classify_job(
+            job_id, project_id=self.project_id, organization_id=self.organization_id
+        )
+        start_time = time.time()
+        while not is_terminal_status(job.status):
+            polling_duration = time.time() - start_time
+            if polling_duration > self.polling_timeout:
+                raise TimeoutError(
+                    f"Job {job_id} timed out after {polling_duration} seconds"
+                )
+            await asyncio.sleep(self.polling_interval)
+            job = await self.client.classifier.get_classify_job(
+                job_id, project_id=self.project_id, organization_id=self.organization_id
+            )
+        return job

py/llama_cloud_services/constants.py

@@ -1 +1,2 @@
 EU_BASE_URL = "https://api.cloud.eu.llamaindex.ai"
+POLLING_TIMEOUT_SECONDS = 300.0

py/llama_cloud_services/extract/extract.py

@@ -33,9 +33,9 @@ from llama_cloud.client import AsyncLlamaCloud
 from llama_cloud.core.api_error import ApiError
 from llama_cloud_services.extract.utils import (
     JSONObjectType,
-    augment_async_errors,
     ExperimentalWarning,
 )
+from llama_cloud_services.utils import augment_async_errors
 from llama_index.core.schema import BaseComponent
 from llama_index.core.async_utils import run_jobs
 from llama_index.core.bridge.pydantic import Field, PrivateAttr

py/llama_cloud_services/extract/utils.py

@@ -1,12 +1,4 @@
-from typing import Any, Dict, List, Union, Generator
-from contextlib import contextmanager
-
-# Asyncio error messages
-nest_asyncio_err = "cannot be called from a running event loop"
-nest_asyncio_msg = (
-    "The event loop is already running. "
-    "Add `import nest_asyncio; nest_asyncio.apply()` to your code to fix this issue."
-)
+from typing import Any, Dict, List, Union
 
 
 def is_jupyter() -> bool:
@@ -19,17 +11,6 @@ def is_jupyter() -> bool:
         return False
 
 
-@contextmanager
-def augment_async_errors() -> Generator[None, None, None]:
-    """Context manager to add helpful information for errors due to nested event loops."""
-    try:
-        yield
-    except RuntimeError as e:
-        if nest_asyncio_err in str(e):
-            raise RuntimeError(nest_asyncio_msg)
-        raise
-
-
 JSONType = Union[Dict[str, Any], List[Any], str, int, float, bool, None]
 JSONObjectType = Dict[str, JSONType]
 

py/llama_cloud_services/files/__init__.py

@@ -0,0 +1,3 @@
+from .client import FileClient
+
+__all__ = ["FileClient"]

py/llama_cloud_services/files/client.py

@@ -0,0 +1,97 @@
+from io import BytesIO
+from typing import BinaryIO
+import os
+from llama_cloud.client import AsyncLlamaCloud
+from llama_cloud.types import File, FileCreate
+from typing import Optional
+
+
+class FileClient:
+    """
+    Higher-level client for interacting with the LlamaCloud Files API.
+    Uses presigned URLs for uploads by default.
+
+    Args:
+        client: The LlamaCloud client to use.
+        project_id: The project ID to use.
+        organization_id: The organization ID to use.
+        use_presigned_url: Whether to use presigned URLs for uploads (set to False when uploading to BYOC deployments).
+    """
+
+    def __init__(
+        self,
+        client: AsyncLlamaCloud,
+        project_id: Optional[str] = None,
+        organization_id: Optional[str] = None,
+        use_presigned_url: bool = True,
+    ):
+        self.client = client
+        self.project_id = project_id
+        self.organization_id = organization_id
+        self.use_presigned_url = use_presigned_url
+
+    async def get_file(self, file_id: str) -> File:
+        return await self.client.files.get_file(
+            file_id, project_id=self.project_id, organization_id=self.organization_id
+        )
+
+    async def read_file_content(self, file_id: str) -> bytes:
+        presigned_url = await self.client.files.read_file_content(
+            file_id,
+            project_id=self.project_id,
+            organization_id=self.organization_id,
+        )
+        httpx_client = self.client._client_wrapper.httpx_client
+        response = await httpx_client.get(presigned_url.url)
+        response.raise_for_status()
+        return response.content
+
+    async def upload_file(
+        self, file_path: str, external_file_id: Optional[str] = None
+    ) -> File:
+        external_file_id = external_file_id or file_path
+        file_size = os.path.getsize(file_path)
+        with open(file_path, "rb") as file:
+            return await self.upload_buffer(file, external_file_id, file_size)
+
+    async def upload_bytes(self, bytes: bytes, external_file_id: str) -> File:
+        return await self.upload_buffer(BytesIO(bytes), external_file_id, len(bytes))
+
+    async def upload_buffer(
+        self,
+        buffer: BinaryIO,
+        external_file_id: str,
+        file_size: int,
+    ) -> File:
+        if self.use_presigned_url:
+            if getattr(buffer, "name", None):
+                name = os.path.basename(str(getattr(buffer, "name", external_file_id)))
+            else:
+                name = external_file_id
+            presigned_url = await self.client.files.generate_presigned_url(
+                project_id=self.project_id,
+                organization_id=self.organization_id,
+                request=FileCreate(
+                    name=name,
+                    external_file_id=external_file_id,
+                    file_size=file_size,
+                ),
+            )
+            httpx_client = self.client._client_wrapper.httpx_client
+            upload_response = await httpx_client.put(
+                presigned_url.url,
+                data=buffer.read(),
+            )
+            upload_response.raise_for_status()
+            return await self.client.files.get_file(
+                presigned_url.file_id,
+                project_id=self.project_id,
+                organization_id=self.organization_id,
+            )
+        else:
+            return await self.client.files.upload_file(
+                upload_file=buffer,
+                external_file_id=external_file_id,
+                project_id=self.project_id,
+                organization_id=self.organization_id,
+            )

py/llama_cloud_services/parse/types.py

@@ -112,7 +112,7 @@ class ChartItem(BaseModel):
 class Page(BaseModel):
     """A page of the document."""
 
-    page: int = Field(description="The page number.")
+    page: int = Field(default=0, description="The page number.")
     text: Optional[str] = Field(default=None, description="The text of the page.")
     md: Optional[str] = Field(default=None, description="The markdown of the page.")
     images: List[ImageItem] = Field(
@@ -153,6 +153,12 @@ class Page(BaseModel):
     noTextContent: bool = Field(
         default=False, description="Whether the page has no text content."
     )
+    isAudioTranscript: bool = Field(
+        default=False, description="Whether the page is an audio transcript."
+    )
+    durationInSeconds: Optional[float] = Field(
+        default=None, description="The duration of the audio transcript in seconds."
+    )
 
 
 class JobResult(BaseModel):

py/llama_cloud_services/utils.py

@@ -1,12 +1,21 @@
 import os
 import importlib.metadata
-
+from contextlib import contextmanager
+from typing import Generator
 import difflib
+from llama_cloud.types import StatusEnum
 import httpx
 import packaging.version
 from pydantic import BaseModel
 from typing import Any, Dict, List, Tuple, Type
 
+# Asyncio error messages
+nest_asyncio_err = "cannot be called from a running event loop"
+nest_asyncio_msg = (
+    "The event loop is already running. "
+    "Add `import nest_asyncio; nest_asyncio.apply()` to your code to fix this issue."
+)
+
 
 def check_extra_params(
     model_cls: Type[BaseModel], data: Dict[str, Any]
@@ -34,6 +43,25 @@ def check_extra_params(
     return extra_params, suggestions
 
 
+def is_terminal_status(status: StatusEnum) -> bool:
+    """
+    Check if a status is terminal, i.e. the job is done and no more updates are expected.
+    Note: this must be updated if the status enum is updated.
+
+    Args:
+        status: The status to check
+
+    Returns:
+        True if the status is terminal, False otherwise
+    """
+    return status in {
+        StatusEnum.SUCCESS,
+        StatusEnum.ERROR,
+        StatusEnum.CANCELLED,
+        StatusEnum.PARTIAL_SUCCESS,
+    }
+
+
 async def check_for_updates(client: httpx.AsyncClient, quiet: bool = True) -> bool:
     """Check if an SDK update is available.
 
@@ -65,3 +93,14 @@ async def check_for_updates(client: httpx.AsyncClient, quiet: bool = True) -> bo
     elif not quiet:
         print(f"{package_name} is up to date")
     return False
+
+
+@contextmanager
+def augment_async_errors() -> Generator[None, None, None]:
+    """Context manager to add helpful information for errors due to nested event loops."""
+    try:
+        yield
+    except RuntimeError as e:
+        if nest_asyncio_err in str(e):
+            raise RuntimeError(nest_asyncio_msg)
+        raise

py/llama_parse/pyproject.toml

@@ -11,13 +11,13 @@ dev = [
 
 [project]
 name = "llama-parse"
-version = "0.6.56"
+version = "0.6.58"
 description = "Parse files into RAG-Optimized formats."
 authors = [{name = "Logan Markewich", email = "logan@llamaindex.ai"}]
 requires-python = ">=3.9,<4.0"
 readme = "README.md"
 license = "MIT"
-dependencies = ["llama-cloud-services>=0.6.56"]
+dependencies = ["llama-cloud-services>=0.6.58"]
 
 [project.scripts]
 llama-parse = "llama_parse.cli.main:parse"

py/pyproject.toml

@@ -12,12 +12,13 @@ dev = [
   "deepdiff>=8.1.1,<9",
   "ipython>=8.12.3,<9",
   "jupyter>=1.1.1,<2",
-  "mypy>=1.14.1,<2"
+  "mypy>=1.14.1,<2",
+  "pydantic-settings>=2.10.1"
 ]
 
 [project]
 name = "llama-cloud-services"
-version = "0.6.56"
+version = "0.6.58"
 description = "Tailored SDK clients for LlamaCloud services."
 authors = [{name = "Logan Markewich", email = "logan@runllama.ai"}]
 requires-python = ">=3.9,<4.0"

py/pytest.ini

@@ -0,0 +1,2 @@
+[pytest]
+asyncio_mode = auto

py/tests/classifier/test_client.py

@@ -0,0 +1,212 @@
+import os
+import pytest
+from llama_cloud.client import AsyncLlamaCloud
+from llama_cloud.types import Project, ClassifierRule, ClassifyJobResults
+from llama_cloud_services.beta.classifier.client import ClassifyClient
+from llama_cloud_services.files.client import FileClient
+from llama_cloud.errors.unprocessable_entity_error import UnprocessableEntityError
+from tests.conftest import EndToEndTestSettings
+import nest_asyncio
+
+# Skip all tests if API key is not set
+pytestmark = pytest.mark.skipif(
+    not os.getenv("LLAMA_CLOUD_API_KEY"), reason="LLAMA_CLOUD_API_KEY not set"
+)
+
+
+@pytest.fixture
+def async_llama_cloud_client(
+    e2e_test_settings: EndToEndTestSettings,
+) -> AsyncLlamaCloud:
+    return AsyncLlamaCloud(
+        token=e2e_test_settings.LLAMA_CLOUD_API_KEY.get_secret_value(),
+        base_url=e2e_test_settings.LLAMA_CLOUD_BASE_URL,
+    )
+
+
+@pytest.fixture
+async def project(
+    async_llama_cloud_client: AsyncLlamaCloud, e2e_test_settings: EndToEndTestSettings
+) -> Project:
+    projects = await async_llama_cloud_client.projects.list_projects(
+        project_name=e2e_test_settings.LLAMA_CLOUD_PROJECT_NAME,
+        organization_id=e2e_test_settings.LLAMA_CLOUD_ORGANIZATION_ID,
+    )
+    assert len(projects) == 1
+    return projects[0]
+
+
+@pytest.fixture
+def classify_client(
+    async_llama_cloud_client: AsyncLlamaCloud, project: Project
+) -> ClassifyClient:
+    return ClassifyClient(
+        async_llama_cloud_client,
+        project_id=project.id,
+        organization_id=project.organization_id,
+        polling_interval=1,
+    )
+
+
+@pytest.fixture
+def file_client(
+    async_llama_cloud_client: AsyncLlamaCloud, project: Project
+) -> FileClient:
+    return FileClient(
+        async_llama_cloud_client,
+        project_id=project.id,
+        organization_id=project.organization_id,
+        use_presigned_url=False,
+    )
+
+
+@pytest.fixture(scope="session", autouse=True)
+def apply_nest_asyncio():
+    nest_asyncio.apply()
+
+
+@pytest.fixture
+def simple_pdf_file_path() -> str:
+    return "tests/test_files/index/Simple PDF Slides.pdf"
+
+
+@pytest.fixture
+def research_paper_path() -> str:
+    return "tests/test_files/attention_is_all_you_need_chart.pdf"
+
+
+@pytest.fixture
+def classification_rules() -> list[ClassifierRule]:
+    return [
+        ClassifierRule(
+            type="number",
+            description="Documents with numbers",
+        ),
+        ClassifierRule(
+            type="research_paper",
+            description="Research papers",
+        ),
+    ]
+
+
+parameterize_sync_and_async = pytest.mark.parametrize("sync", [True, False])
+
+
+@parameterize_sync_and_async
+@pytest.mark.asyncio
+async def test_classify_file_ids(
+    classify_client: ClassifyClient,
+    file_client: FileClient,
+    simple_pdf_file_path: str,
+    research_paper_path: str,
+    classification_rules: list[ClassifierRule],
+    sync: bool,
+):
+    """Test classifying files by their IDs"""
+    # Upload test files first to get their IDs
+    pdf_file = await file_client.upload_file(simple_pdf_file_path)
+    research_paper_file = await file_client.upload_file(research_paper_path)
+
+    # Classify the uploaded files
+    if sync:
+        results = classify_client.classify_file_ids(
+            rules=classification_rules, file_ids=[pdf_file.id, research_paper_file.id]
+        )
+    else:
+        results = await classify_client.aclassify_file_ids(
+            rules=classification_rules, file_ids=[pdf_file.id, research_paper_file.id]
+        )
+
+    assert isinstance(results, ClassifyJobResults)
+    assert len(results.items) == 2
+
+    file_id_to_expected_type = {
+        pdf_file.id: "number",
+        research_paper_file.id: "research_paper",
+    }
+    # Verify each file got classified
+    for item in results.items:
+        expected_type = file_id_to_expected_type[item.file_id]
+        assert item.result.type == expected_type
+
+
+@parameterize_sync_and_async
+@pytest.mark.asyncio
+async def test_classify_file_path(
+    classify_client: ClassifyClient,
+    simple_pdf_file_path: str,
+    classification_rules: list[ClassifierRule],
+    sync: bool,
+):
+    """Test classifying a single file by path"""
+    # Classify the file
+    if sync:
+        results = classify_client.classify_file_path(
+            rules=classification_rules, file_input_path=simple_pdf_file_path
+        )
+    else:
+        results = await classify_client.aclassify_file_path(
+            rules=classification_rules, file_input_path=simple_pdf_file_path
+        )
+
+    assert isinstance(results, ClassifyJobResults)
+    assert len(results.items) == 1
+
+    # Verify the file got classified
+    item = results.items[0]
+    assert item.result.type == "number"
+
+
+@parameterize_sync_and_async
+@pytest.mark.asyncio
+async def test_classify_file_paths(
+    classify_client: ClassifyClient,
+    file_client: FileClient,
+    simple_pdf_file_path: str,
+    research_paper_path: str,
+    classification_rules: list[ClassifierRule],
+    sync: bool,
+):
+    """Test classifying multiple files by paths"""
+    # Classify all test files
+    if sync:
+        results = classify_client.classify_file_paths(
+            rules=classification_rules,
+            file_input_paths=[simple_pdf_file_path, research_paper_path],
+        )
+    else:
+        results = await classify_client.aclassify_file_paths(
+            rules=classification_rules,
+            file_input_paths=[simple_pdf_file_path, research_paper_path],
+        )
+
+    assert isinstance(results, ClassifyJobResults)
+    assert len(results.items) == 2
+
+    file_name_to_expected_type = {
+        os.path.basename(simple_pdf_file_path): "number",
+        os.path.basename(research_paper_path): "research_paper",
+    }
+    # Verify each file got classified
+    for item in results.items:
+        file = await file_client.get_file(item.file_id)
+        expected_type = file_name_to_expected_type[file.name]
+        assert item.result.type == expected_type
+
+
+@parameterize_sync_and_async
+@pytest.mark.asyncio
+async def test_classify_empty_file_list(
+    classify_client: ClassifyClient,
+    classification_rules: list[ClassifierRule],
+    sync: bool,
+):
+    """Test classifying an empty list of files"""
+    # This should throw an error
+    with pytest.raises(UnprocessableEntityError):
+        if sync:
+            classify_client.classify_file_ids(rules=classification_rules, file_ids=[])
+        else:
+            await classify_client.aclassify_file_ids(
+                rules=classification_rules, file_ids=[]
+            )

py/tests/conftest.py

@@ -0,0 +1,26 @@
+import pytest
+from llama_index.core.constants import DEFAULT_BASE_URL
+from pydantic import Field, SecretStr
+from pydantic_settings import BaseSettings
+
+
+class EndToEndTestSettings(BaseSettings):
+    LLAMA_CLOUD_BASE_URL: str = Field(
+        description="The base URL of the LlamaCloud API", default=DEFAULT_BASE_URL
+    )
+    LLAMA_CLOUD_API_KEY: SecretStr = Field(
+        description="The API key for the LlamaCloud API"
+    )
+    LLAMA_CLOUD_ORGANIZATION_ID: str | None = Field(
+        description="The organization ID for the LlamaCloud API",
+        default=None,
+    )
+    LLAMA_CLOUD_PROJECT_NAME: str = Field(
+        description="The project name for the LlamaCloud API",
+        default="framework_integration_test",
+    )
+
+
+@pytest.fixture
+def e2e_test_settings() -> EndToEndTestSettings:
+    return EndToEndTestSettings()

py/tests/index/test_index.py

@@ -27,8 +27,6 @@ api_key = os.environ.get("LLAMA_CLOUD_API_KEY", None)
 organization_id = os.environ.get("LLAMA_CLOUD_ORGANIZATION_ID", None)
 project_name = os.environ.get("LLAMA_CLOUD_PROJECT_NAME", "framework_integration_test")
 
-print("api-key", api_key, "base-url", base_url)
-
 
 @pytest.fixture()
 def remote_file() -> Tuple[str, str]:

py/tests/parse/test_llama_parse.py

@@ -202,3 +202,12 @@ async def test_get_result(markdown_parser: LlamaParse) -> None:
     result = await markdown_parser.aget_result(expected.job_id)
     assert result.job_id == expected.job_id
     assert len(result.pages) == len(expected.pages)
+
+
+@pytest.mark.asyncio
+async def test_parse_audio() -> None:
+    parser = LlamaParse()
+    filepath = "tests/test_files/hello_world.m4a"
+
+    result = await parser.aparse(filepath)
+    assert result.job_id is not None

py/tests/test_files_client.py

@@ -0,0 +1,147 @@
+import os
+import pytest
+from io import BytesIO
+from llama_cloud.client import AsyncLlamaCloud
+from llama_cloud.types import Project, File
+from llama_cloud_services.files.client import FileClient
+from tests.conftest import EndToEndTestSettings
+
+
+@pytest.fixture
+async def llama_cloud_client(
+    e2e_test_settings: EndToEndTestSettings,
+) -> AsyncLlamaCloud:
+    return AsyncLlamaCloud(
+        token=e2e_test_settings.LLAMA_CLOUD_API_KEY.get_secret_value(),
+        base_url=e2e_test_settings.LLAMA_CLOUD_BASE_URL,
+    )
+
+
+@pytest.fixture
+async def project(
+    llama_cloud_client: AsyncLlamaCloud, e2e_test_settings: EndToEndTestSettings
+) -> Project:
+    projects = await llama_cloud_client.projects.list_projects(
+        project_name=e2e_test_settings.LLAMA_CLOUD_PROJECT_NAME,
+        organization_id=e2e_test_settings.LLAMA_CLOUD_ORGANIZATION_ID,
+    )
+    assert len(projects) == 1
+    return projects[0]
+
+
+@pytest.fixture
+async def file_client(
+    llama_cloud_client: AsyncLlamaCloud, project: Project, use_presigned_url: bool
+) -> FileClient:
+    return FileClient(
+        llama_cloud_client,
+        project_id=project.id,
+        organization_id=project.organization_id,
+        use_presigned_url=use_presigned_url,
+    )
+
+
+@pytest.fixture
+def test_file() -> str:
+    return "tests/test_files/index/Simple PDF Slides.pdf"
+
+
+parametrize_use_presigned_url = pytest.mark.parametrize(
+    "use_presigned_url", [True, False]
+)
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_upload_file_from_path(file_client: FileClient, test_file: str):
+    """Test uploading a file from file path"""
+    external_file_id = f"test_upload_path_{os.getpid()}"
+    uploaded_file = await file_client.upload_file(test_file, external_file_id)
+
+    assert isinstance(uploaded_file, File)
+    expected_name = os.path.basename(test_file)
+    assert uploaded_file.name == expected_name
+    assert uploaded_file.external_file_id == external_file_id
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_upload_bytes(
+    file_client: FileClient, test_file: str, use_presigned_url: bool
+):
+    """Test uploading a file from bytes"""
+    # Read file as bytes
+    with open(test_file, "rb") as f:
+        file_bytes = f.read()
+
+    external_file_id = f"test_upload_bytes_{os.getpid()}"
+    uploaded_file = await file_client.upload_bytes(file_bytes, external_file_id)
+
+    assert isinstance(uploaded_file, File)
+    expected_name = external_file_id if use_presigned_url else "upload"
+    assert uploaded_file.name == expected_name
+    assert uploaded_file.external_file_id == external_file_id
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_upload_buffer(
+    file_client: FileClient, test_file: str, use_presigned_url: bool
+):
+    """Test uploading a file from buffer"""
+    # Read file as bytes and create buffer
+    with open(test_file, "rb") as f:
+        file_bytes = f.read()
+
+    buffer = BytesIO(file_bytes)
+    file_size = len(file_bytes)
+    external_file_id = f"test_upload_buffer_{os.getpid()}"
+
+    uploaded_file = await file_client.upload_buffer(buffer, external_file_id, file_size)
+
+    assert isinstance(uploaded_file, File)
+    expected_name = external_file_id if use_presigned_url else "upload"
+    assert uploaded_file.name == expected_name
+    assert uploaded_file.external_file_id == external_file_id
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_get_file(file_client: FileClient, test_file: str):
+    """Test retrieving a file by ID"""
+    # Upload a file first
+    external_file_id = f"test_get_file_{os.getpid()}"
+    uploaded_file = await file_client.upload_file(test_file, external_file_id)
+
+    # Retrieve the file by ID
+    retrieved_file = await file_client.get_file(uploaded_file.id)
+
+    assert isinstance(retrieved_file, File)
+    assert retrieved_file == uploaded_file
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_upload_with_default_external_id(file_client: FileClient, test_file: str):
+    """Test uploading file with default external_file_id"""
+    # Upload file without specifying external_file_id
+    uploaded_file = await file_client.upload_file(test_file)
+
+    assert isinstance(uploaded_file, File)
+    assert uploaded_file.name == os.path.basename(test_file)
+    assert uploaded_file.external_file_id == test_file
+
+
+@parametrize_use_presigned_url
+@pytest.mark.asyncio
+async def test_read_file_content(file_client: FileClient, test_file: str):
+    """Test reading a file content"""
+    # Upload a file first
+    external_file_id = f"test_read_file_content_{os.getpid()}"
+    uploaded_file = await file_client.upload_file(test_file, external_file_id)
+
+    # Read the file content
+    file_content = await file_client.read_file_content(uploaded_file.id)
+    with open(test_file, "rb") as f:
+        expected_file_content = f.read()
+    assert file_content == expected_file_content

py/unit_tests/beta/agent/test_agent_data_schema.py

@@ -1,15 +1,18 @@
 from datetime import datetime
-from typing import Any, Dict
+import json
+from pathlib import Path
+from typing import Any, Dict, Optional
 
 import pytest
 from llama_cloud import ExtractRun, File
 from llama_cloud.types.agent_data import AgentData
 from llama_cloud.types.aggregate_group import AggregateGroup
-from pydantic import BaseModel, ValidationError
+from pydantic import BaseModel, Field, ValidationError
 
 from llama_cloud_services.beta.agent_data.schema import (
     ExtractedData,
     ExtractedFieldMetadata,
+    FieldCitation,
     InvalidExtractionData,
     TypedAgentData,
     TypedAggregateGroup,
@@ -81,8 +84,12 @@ def test_extracted_data_create_method():
 
     # Test with custom values using ExtractedFieldMetadata
     field_metadata = {
-        "name": ExtractedFieldMetadata(confidence=0.99, page_number=1),
-        "age": ExtractedFieldMetadata(confidence=0.85, page_number=1),
+        "name": ExtractedFieldMetadata(
+            confidence=0.99, citation=[FieldCitation(page=1)]
+        ),
+        "age": ExtractedFieldMetadata(
+            confidence=0.85, citation=[FieldCitation(page=1)]
+        ),
     }
     extracted_custom = ExtractedData.create(
         person, status="accepted", field_metadata=field_metadata
@@ -254,14 +261,16 @@ def test_parse_extracted_field_metadata():
     # name should have parsed citation data
     assert isinstance(result["name"], ExtractedFieldMetadata)
     assert result["name"].confidence == 0.95
-    assert result["name"].page_number == 1
-    assert result["name"].matching_text == "John Smith"
+    assert result["name"].citation == [
+        FieldCitation(page=1, matching_text="John Smith")
+    ]
 
     # age should handle float page number
     assert isinstance(result["age"], ExtractedFieldMetadata)
     assert result["age"].confidence == 0.87
-    assert result["age"].page_number == 2  # Should be converted to int
-    assert result["age"].matching_text == "25 years old"
+    assert result["age"].citation == [
+        FieldCitation(page=2, matching_text="25 years old")
+    ]
 
     # email should handle empty citations
     assert isinstance(result["email"], ExtractedFieldMetadata)
@@ -327,30 +336,38 @@ def test_parse_extracted_field_metadata_complex():
             reasoning="Combined key parametrics and construction from the datasheet for a structured title.",
             confidence=0.9470628580889779,
             extraction_confidence=0.9470628580889779,
-            page_number=1,
-            matching_text="PHE844/F844, Film, Metallized Polypropylene, Safety, 0.47 uF",
+            citation=[
+                FieldCitation(
+                    page=1,
+                    matching_text="PHE844/F844, Film, Metallized Polypropylene, Safety, 0.47 uF",
+                )
+            ],
         ),
         "manufacturer": ExtractedFieldMetadata(
             reasoning="VERBATIM EXTRACTION",
             confidence=0.9997446550976602,
             extraction_confidence=0.9997446550976602,
-            page_number=1,
-            matching_text="YAGEO KEMET",
+            citation=[FieldCitation(page=1, matching_text="YAGEO KEMET")],
         ),
         "features": [
             ExtractedFieldMetadata(
                 reasoning="VERBATIM EXTRACTION",
                 confidence=0.9999308195540074,
                 extraction_confidence=0.9999308195540074,
-                page_number=1,
-                matching_text="Features</td><td>EMI Safety",
+                citation=[
+                    FieldCitation(
+                        page=1,
+                        matching_text="Features</td><td>EMI Safety",
+                    )
+                ],
             ),
             ExtractedFieldMetadata(
                 reasoning="VERBATIM EXTRACTION",
                 confidence=0.8642493886452225,
                 extraction_confidence=0.8642493886452225,
-                page_number=1,
-                matching_text="THB Performance</td><td>Yes",
+                citation=[
+                    FieldCitation(page=1, matching_text="THB Performance</td><td>Yes")
+                ],
             ),
         ],
         "dimensions": {
@@ -358,15 +375,13 @@ def test_parse_extracted_field_metadata_complex():
                 reasoning="VERBATIM EXTRACTION",
                 confidence=0.8986941382802304,
                 extraction_confidence=0.8986941382802304,
-                page_number=1,
-                matching_text="L</td><td>41mm MAX",
+                citation=[FieldCitation(page=1, matching_text="L</td><td>41mm MAX")],
             ),
             "width": ExtractedFieldMetadata(
                 reasoning="VERBATIM EXTRACTION",
                 confidence=0.9999377974447091,
                 extraction_confidence=0.9999377974447091,
-                page_number=1,
-                matching_text="T</td><td>13mm MAX",
+                citation=[FieldCitation(page=1, matching_text="T</td><td>13mm MAX")],
             ),
         },
     }
@@ -450,8 +465,9 @@ def test_extracted_data_from_extraction_result_success():
     # Verify field metadata was parsed
     assert isinstance(extracted.field_metadata["name"], ExtractedFieldMetadata)
     assert extracted.field_metadata["name"].confidence == 0.95
-    assert extracted.field_metadata["name"].page_number == 1
-    assert extracted.field_metadata["name"].matching_text == "John Doe"
+    assert extracted.field_metadata["name"].citation == [
+        FieldCitation(page=1, matching_text="John Doe")
+    ]
 
     # Verify overall confidence was calculated
     expected_confidence = (0.95 + 0.87 + 0.92) / 3
@@ -523,3 +539,54 @@ def test_extracted_data_from_extraction_result_invalid_data():
     assert isinstance(invalid_data.field_metadata["name"], ExtractedFieldMetadata)
     assert invalid_data.field_metadata["name"].confidence == 0.9
     assert invalid_data.overall_confidence == 0.9
+
+
+class Dimensions(BaseModel):
+    length: Optional[str] = Field(
+        None, description="Length in mm (Size, Longest Side, L)"
+    )
+    width: Optional[str] = Field(
+        None, description="Width in mm (Breadth, Side Width, W)"
+    )
+    height: Optional[str] = Field(
+        None, description="Height in mm (Thickness, Vertical Size, H)"
+    )
+    diameter: Optional[str] = Field(
+        None,
+        description="Diameter in mm (for radial or cylindrical types) (Outer Diameter, dt, OD, D, d<sub>t</sub>)",
+    )
+    lead_spacing: Optional[str] = Field(
+        None, description="Lead spacing in mm (Pin Pitch, Terminal Gap, LS)"
+    )
+
+
+class Capacitor(BaseModel):
+    dimensions: Optional[Dimensions] = None
+
+
+def test_full_parse_nested_dimensions():
+    with open(Path(__file__).parent.parent.parent / "data" / "capacitor.json") as f:
+        data = json.load(f)
+    result = ExtractedData.from_extraction_result(ExtractRun.parse_obj(data), Capacitor)
+    expected = {
+        "dimensions": {
+            "diameter": ExtractedFieldMetadata(
+                reasoning="VERBATIM EXTRACTION",
+                confidence=1.0,
+                extraction_confidence=1.0,
+            ),
+            "lead_spacing": ExtractedFieldMetadata(
+                reasoning="VERBATIM EXTRACTION",
+                confidence=0.9999999031936799,
+                extraction_confidence=0.9999999031936799,
+            ),
+            "length": ExtractedFieldMetadata(
+                reasoning="VERBATIM EXTRACTION",
+                confidence=0.9999968039036192,
+                extraction_confidence=0.9999968039036192,
+            ),
+        }
+    }
+    assert result.field_metadata == expected
+    parsed = ExtractedData.model_validate_json(result.model_dump_json())
+    assert parsed.field_metadata == expected

py/unit_tests/data/capacitor.json

@@ -0,0 +1,116 @@
+{
+  "id": "de058dda-6ca7-4eea-a426-da802f84f971",
+  "created_at": "2025-08-13T15:45:39.286921Z",
+  "updated_at": "2025-08-13T15:47:04.878069Z",
+  "extraction_agent_id": "e834e99f-1f35-4748-b82f-03de4bd07ca6",
+  "data_schema": {
+    "additionalProperties": false,
+    "properties": {
+      "dimensions": {
+        "anyOf": [
+          {
+            "additionalProperties": false,
+            "properties": {
+              "length": {
+                "anyOf": [{ "type": "string" }, { "type": "null" }],
+                "description": "Length in mm (Size, Longest Side, L)"
+              },
+              "width": {
+                "anyOf": [{ "type": "string" }, { "type": "null" }],
+                "description": "Width in mm (Breadth, Side Width, W)"
+              },
+              "height": {
+                "anyOf": [{ "type": "string" }, { "type": "null" }],
+                "description": "Height in mm (Thickness, Vertical Size, H)"
+              },
+              "diameter": {
+                "anyOf": [{ "type": "string" }, { "type": "null" }],
+                "description": "Diameter in mm (for radial or cylindrical types) (Outer Diameter, OD, D)"
+              },
+              "lead_spacing": {
+                "anyOf": [{ "type": "string" }, { "type": "null" }],
+                "description": "Lead spacing in mm (Pin Pitch, Terminal Gap, LS)"
+              }
+            },
+            "required": [
+              "length",
+              "width",
+              "height",
+              "diameter",
+              "lead_spacing"
+            ],
+            "type": "object"
+          },
+          { "type": "null" }
+        ]
+      }
+    },
+    "required": ["dimensions"],
+    "type": "object"
+  },
+  "config": {
+    "priority": null,
+    "extraction_target": "PER_DOC",
+    "extraction_mode": "PREMIUM",
+    "multimodal_fast_mode": false,
+    "system_prompt": "",
+    "use_reasoning": true,
+    "cite_sources": false,
+    "confidence_scores": true,
+    "chunk_mode": "PAGE",
+    "high_resolution_mode": false,
+    "invalidate_cache": false,
+    "page_range": null
+  },
+  "file": {
+    "id": "9233cc2b-00e3-4ddd-b426-b8b59357d4cb",
+    "created_at": "2025-08-12T18:31:54.269440Z",
+    "updated_at": "2025-08-13T15:45:39.064906Z",
+    "name": "document (2).pdf.txt",
+    "external_file_id": "document (2).pdf.txt",
+    "file_size": 2562,
+    "file_type": "txt",
+    "project_id": "77bdc79f-fb69-49ae-a783-fcc573eec7ce",
+    "last_modified_at": "2025-08-13T15:45:39Z",
+    "resource_info": {
+      "file_size": 2562,
+      "last_modified_at": "2025-08-13T15:45:39"
+    },
+    "permission_info": null,
+    "data_source_id": null
+  },
+  "status": "SUCCESS",
+  "error": null,
+  "job_id": "5bb8a583-366c-416c-ba55-4f5724fef9a9",
+  "data": {
+    "dimensions": {
+      "length": "82 mm",
+      "width": null,
+      "height": null,
+      "diameter": "35 mm",
+      "lead_spacing": "6.0 mm"
+    }
+  },
+  "extraction_metadata": {
+    "field_metadata": {
+      "dimensions": {
+        "length": {
+          "extraction_confidence": 0.9999968039036192,
+          "confidence": 0.9999968039036192
+        },
+        "diameter": { "extraction_confidence": 1.0, "confidence": 1.0 },
+        "lead_spacing": {
+          "extraction_confidence": 0.9999999031936799,
+          "confidence": 0.9999999031936799
+        },
+        "reasoning": "VERBATIM EXTRACTION"
+      }
+    },
+    "usage": {
+      "num_pages_extracted": 2,
+      "num_document_tokens": 1034,
+      "num_output_tokens": 3440
+    }
+  },
+  "from_ui": false
+}

py/uv.lock

@@ -1587,7 +1587,7 @@ wheels = [
 
 [[package]]
 name = "llama-cloud-services"
-version = "0.6.56"
+version = "0.6.58"
 source = { editable = "." }
 dependencies = [
     { name = "click", version = "8.1.8", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.10'" },
@@ -1612,6 +1612,7 @@ dev = [
     { name = "jupyter" },
     { name = "mypy" },
     { name = "pre-commit" },
+    { name = "pydantic-settings" },
     { name = "pytest" },
     { name = "pytest-asyncio" },
 ]
@@ -1638,6 +1639,7 @@ dev = [
     { name = "jupyter", specifier = ">=1.1.1,<2" },
     { name = "mypy", specifier = ">=1.14.1,<2" },
     { name = "pre-commit", specifier = "==3.2.0" },
+    { name = "pydantic-settings", specifier = ">=2.10.1" },
     { name = "pytest", specifier = ">=8.0.0,<9" },
     { name = "pytest-asyncio" },
 ]
@@ -2855,6 +2857,20 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d4/29/3cade8a924a61f60ccfa10842f75eb12787e1440e2b8660ceffeb26685e7/pydantic_core-2.33.2-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:2807668ba86cb38c6817ad9bc66215ab8584d1d304030ce4f0887336f28a5e27", size = 2066661, upload-time = "2025-04-23T18:33:49.995Z" },
 ]
 
+[[package]]
+name = "pydantic-settings"
+version = "2.10.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pydantic" },
+    { name = "python-dotenv" },
+    { name = "typing-inspection" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/68/85/1ea668bbab3c50071ca613c6ab30047fb36ab0da1b92fa8f17bbc38fd36c/pydantic_settings-2.10.1.tar.gz", hash = "sha256:06f0062169818d0f5524420a360d632d5857b83cffd4d42fe29597807a1614ee", size = 172583, upload-time = "2025-06-24T13:26:46.841Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/58/f0/427018098906416f580e3cf1366d3b1abfb408a0652e9f31600c24a1903c/pydantic_settings-2.10.1-py3-none-any.whl", hash = "sha256:a60952460b99cf661dc25c29c0ef171721f98bfcb52ef8d9ea4c943d7c8cc796", size = 45235, upload-time = "2025-06-24T13:26:45.485Z" },
+]
+
 [[package]]
 name = "pygments"
 version = "2.19.2"

ts/llama_cloud_services/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llama-cloud-services",
-  "version": "0.3.1",
+  "version": "0.3.2",
   "type": "module",
   "license": "MIT",
   "scripts": {

ts/llama_cloud_services/src/beta/agent/types.ts

@@ -38,8 +38,12 @@ export interface ExtractedFieldMetadata {
   confidence?: number;
   /** The confidence score for the field based on the extracted text only */
   extraction_confidence?: number;
+  citation: FieldCitation[];
+}
+
+export interface FieldCitation {
   /** The page number that the field occurred on */
-  page_number?: number;
+  page?: number;
   /** The original text this field's value was derived from */
   matching_text?: string;
 }