v0.1.0

Initial release. Targets Compliance API spec revision Rev K
(2026-05-04).

Added

• Public clients — ComplianceClient (sync) and
  AsyncComplianceClient (async). Identical resource surface on both;
  swap one for the other and add await to switch styles.
• Resource groups covering every Rev K endpoint:
  • activities — cursor-paginated Activity Feed list / iter.
  • organizations — unpaginated org list, plus paginated users
    per org.
  • projects — list, get (detail), delete, and attachment listing.
  • project_documents — fetch document text content and delete.
  • chats — cursor list of chats, combined chat-with-messages
    fetch, message iteration, and delete.
  • files — user-uploaded file metadata, download, delete.
  • generated_files — assistant tool-use outputs (download only;
    not deletable per spec).
  • artifacts — versioned text artifacts (download only).
  • roles — org-scoped role list, fetch, and permissions list.
  • groups — group list, fetch, and member list.
• Pagination — CursorPage[T] (used by the Activity Feed, Chats,
  Messages) and OffsetPage[T] (everything else). Every paginated
  resource exposes both .list() (one page) and .iter()
  (auto-paginate) methods.
• Downloads — three resource groups (files, generated_files,
  artifacts) share the same trio:
  • .download(id) — eager, bounded by max_download_bytes (default
    100 MiB). Raises FileTooLargeError when the cap is exceeded.
  • .download_to_file(id, path) — streamed to disk, unbounded.
  • .download_stream(id) — yields chunks for caller-managed
    streaming.
• Typed exception hierarchy rooted at ComplianceClientError.
  HTTP failures under APIError map every documented status code to
  a typed subclass: BadRequestError, InvalidAPIKeyError /
  InsufficientScopeError (a best-effort split of 401),
  PermissionDeniedError, NotFoundError, ConflictError,
  RateLimitError (carrying retry_after), APIStatusError,
  InternalServerError. Transport-level failures live under
  APIConnectionError / APITimeoutError. Every error carries
  status_code, request_id, error_type, error_message, and
  the raw response body.
• Resilience — exponential-backoff retry on 429 / 5xx / connect
  errors with Retry-After honoured, plus a client-side
  sliding-window rate limiter sized to the server's 600 RPM cap.
  Both tunable / disable-able via max_retries and
  rate_limit_rpm.
• Typed responses — every response is a plain dataclass.
  Activity-type-specific fields and any future-spec additions are
  preserved in an extra: dict so the SDK does not break when the
  spec grows.
• Runnable examples — examples/activity_audit.py,
  examples/ediscovery_export.py, and examples/file_pull.py
  demonstrate the spec's headline compliance use cases end-to-end.
• Documentation site at
  papermtn.github.io/claude-compliance-sdk.