- Accept the API URL the Transform Platform hands you as
server_url. The app's API Keys page and the docs give you https://platform-api.transform.unstructured.io/api/v1, which works with curl but 404'd every Platform call in the SDK: the URL cleaner only stripped a path for unstructuredapp.io hosts, so the /api/v1 survived and the operation's own /api/v1/jobs/ was appended on top, producing /api/v1/api/v1/jobs/. Hosts under unstructured.io are now recognized too, and are matched on domain boundaries so a lookalike host like unstructuredapp.io.example.com keeps its path and scheme. A server_url passed to an individual operation is cleaned as well — previously only the client-level URL was, so client.jobs.list_jobs(request={}, server_url=...) still 404'd.
- Fail loudly when a split-PDF chunk returns HTTP 200 with an empty JSON body in NDJSON elements-file mode. The chunk used to be logged and skipped, so the combined
elements_file was silently short by those pages while the call still returned 200 — and split_pdf_allow_failed=False did not catch it, because an empty 200 counts as a successful chunk. Recombination now raises EmptyChunkResponseError (a ValueError), matching the buffered path, which raises JSONDecodeError on the same response. Emptiness is judged against the chunk's own Content-Type: JSON has no empty document (a chunk with no elements is []), while application/x-ndjson encodes zero records as zero lines, so an empty NDJSON chunk is well formed and still contributes nothing.
- Add an NDJSON elements-file mode to
partition(). Pass accept_header_override=PartitionAcceptEnum.APPLICATION_X_NDJSON to get PartitionResponse.elements_file — a path to an NDJSON file with one element per line — instead of PartitionResponse.elements. On the split-PDF path the per-chunk temp files are concatenated on disk rather than parsed, flattened, re-serialized with json.dumps and re-parsed by the SDK, which held four copies of the document in memory at once; peak memory becomes roughly one chunk instead of the whole document. elements_file is set for every input, including ones that are not split and responses from a server that ignores the Accept header, so callers need only one code path — but the memory saving itself applies only to split PDFs (a PDF of more than two pages, with split_pdf_page=True). The caller owns the returned file and is responsible for deleting it. Requesting application/json (the default) is unchanged.
- Make the split-PDF
httpx.AsyncClient connection-pool limits configurable via env vars: UNSTRUCTURED_CLIENT_MAX_CONNECTIONS (default 100), UNSTRUCTURED_CLIENT_MAX_KEEPALIVE_CONNECTIONS (default 20), and UNSTRUCTURED_CLIENT_KEEPALIVE_EXPIRY (default 5.0s). Defaults match httpx, so behavior is unchanged unless set. Useful when deploying behind a connect-time-only load balancer (e.g. Kubernetes ClusterIP without a mesh) where shorter keepalives force connections to redistribute across backend pods.
- Honor the standard
SSL_CERT_FILE / REQUESTS_CA_BUNDLE env vars to point the split-PDF httpx.AsyncClient at a custom trust store, so a single env-var setting applies uniformly across Python tooling.
- Add
UNSTRUCTURED_CLIENT_TLS_CLIENT_CERT and UNSTRUCTURED_CLIENT_TLS_CLIENT_KEY env vars to wire an mTLS client certificate into the split-PDF httpx.AsyncClient (single PEM, or split cert + key files).
- Extend the split-PDF
event=plan_created log to include the resolved pool limits and trust-store / mTLS mode so the active config is visible in production logs.
- Add
min_attempts and absolute_max_elapsed_time_ms fields to BackoffStrategy. min_attempts is the minimum number of retry attempts that must fire before max_elapsed_time is honored; defaults to 0 (preserves existing behavior). absolute_max_elapsed_time_ms caps when a new retry can start (does not interrupt in-flight requests); defaults to None. Together these close a short-circuit where a single slow first attempt could exhaust the retry budget before any retry fired.
- Removed deprecated connector config models from the SDK (e.g.
S3SourceConnectorConfig, AzureDestinationConnectorConfig). Pass connector configs as plain dicts with arbitrary fields. The SDK is no longer coupled to backend connector schemas — new fields work without an SDK upgrade.
- Route split-PDF
partition_async() result collection through awaited async hook dispatch instead of creating a nested event loop in a worker thread.
Sync-only hooks on the async path now run on a worker thread, so hook code that depends on event-loop-thread contextvars or thread-local state should pass that state explicitly.
- Add cancellation cleanup for in-flight split-PDF chunk tasks and preserve existing sync
partition() split-PDF behavior with lazy executor creation.
- Switch PyPI publishing to GitHub trusted publishing so releases can publish via OIDC without a long-lived
PYPI_TOKEN secret.
- Align release automation, package metadata, and generator config on
0.43.2 for the trusted-publishing release flow.
- Add split-PDF observability with operation-aware batch planning, timeout, cancellation, and completion logs.
- Make long-running integration tests stream live progress, timings, and backend failure context for split and single partition phases.
- Preserve chunk-local transport retries for split-PDF execution even when SDK-level retries disable connection-error retries for top-level requests.
- Harden split-PDF timeout and cleanup paths against closed event loops and cancelled chunk tasks.
- Stabilize
hi_res split integration coverage by using a smaller derived multi-page fixture instead of the flaky full layout-parser-paper.pdf path for equivalence and caching checks.
- Retry on all
httpx.TransportError subclasses (including ReadError, WriteError, ConnectError, RemoteProtocolError, and all timeout types) when retry_connection_errors=True. Previously only ConnectError, RemoteProtocolError, and TimeoutException were retried — ReadError (TCP connection reset mid-response) was treated as permanent.
- Retry on
httpx.RemoteProtocolError (e.g. "Server disconnected without sending a response") when retry_connection_errors=True. Previously, mid-request server crashes were treated as permanent errors and not retried.
- Support for on-demand jobs via CreateJob API
- New Read-only APIs GetTemplate and ListTemplates
- Bump dependencies to account for vulnerabilities in pypdf < 6.1.3
- Enable arbitrary dictionary inputs for
CreateSourceConnectorConfig and CreateDestinationConnectorConfig. This decouples us from the backend schemas. Users can send new connector config fields without having to upgrade their client.
- Enable arbitrary inputs for
SourceConnectorType and DestinationConnectorType. This lets the client support new connector types without having to upgrade.
- potential issue referencing models before declaration (commit by @mfbx9da4)
- Fix some environments failing to split pdfs with
Can't patch loop of type <class 'uvloop.Loop'>, remove usage of nest-asyncio
- Remove some operations under
client.users that are not fully ready yet
- Provide a base
UnstructuredClientError to capture every error raised by the SDK. Note that some exceptions such as SDKError now have more information in the message field. This will impact any users who rely on string matching in their error handling.
- Improve PDF validation error handling by introducing FileValidationError base class for better error abstraction
- Replace RequestError with PDFValidationError for invalid PDF files to provide more accurate error context
- Throws appropriate error message in case the given PDF file is invalid (corrupted or encrypted).
- Add Unstructured Platform APIs to manage source and destination connectors, workflows, and workflow runs
WARNING: This is a breaking change for the use of non-default
server_url settings in the client usage.
To set the custom URL for the client, use the the server_url parameter in a given operation:
elements = client.general.partition(
request=operations.PartitionRequest(
partition_parameters=shared.PartitionParameters(
files=shared.Files(
content=doc_file,
file_name="your_document.pdf",
),
strategy=shared.Strategy.FAST,
)
),
server_url="your_server_url",
)
- Use the configured server_url for our split page "dummy" request
- Switch to a httpx based client instead of requests
- Switch to poetry for dependency management
- Add client side parameter checking via Pydantic or TypedDict interfaces
- Add
partition_async for a non blocking alternative to partition
- Address some asyncio based errors in pdf splitting logic