kurious 0.8.4__tar.gz

kurious-0.8.4/.gitignore

@@ -0,0 +1,277 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+#poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+#pdm.lock
+#pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+#pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Per-environment secrets (real values). The .example variant is committed.
+deployment/environments/*/env.secrets.yaml
+
+# scripts/lib/ holds shared shell + python helpers; the `lib/` rule earlier
+# (Python packaging) would otherwise ignore it.
+!scripts/lib/
+!scripts/lib/**
+
+# Rendered templates produced by `python3 scripts/lib/load_env_config.py render ...`
+# — the .tmpl source is committed; the rendered output is ephemeral.
+*.rendered.yaml
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Abstra
+# Abstra is an AI-powered process automation framework.
+# Ignore directories containing user credentials, local state, and settings.
+# Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#  Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#  that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#  and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#  you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Cursor
+#  Cursor is an AI-powered code editor. `.cursorignore` specifies files/directories to
+#  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
+#  refer to https://docs.cursor.com/context/ignore-files
+.cursorignore
+.cursorindexingignore
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+temp/*.csv
+temp/*.json
+
+data 
+*.out 
+
+sonar-scan.sh
+sonar-project.properties
+# Per-service sonar configs needed by ci-*.yml per-service scans — whitelisted.
+!services/*/sonar-project.properties
+.scannerwork.sonar-token
+.sonar-token
+reports/*.md
+reports/*.json
+
+gh-issue-agent/
+.hitl_state/
+
+# AgentOS — local-only skill overrides (never commit)
+.claude/*.local.*
+.claude/**/*.local.*
+services/kurious-engine/credentials
+.claude/scheduled_tasks.lock
+
+# SonarQube local-scan artifacts
+.scannerwork/
+sonar-project.local.properties
+.DS_Store
+
+# Internal planning docs — not for repo
+docs/6 weeks Sprint prioritisation.xlsx
+docs/sprint-9-12-ml-mlops-plan.md
+
+services/kurious-engine/cov_annotate/
+services/kurious-worker/cov_annotate/
+services/users-service/cov_annotate/
+services/metering-service/cov_annotate/
+cov_annotate_worker
+
+# bug-issues-digest skill: root-level run output (sample lives in docs/progress-tracking/)
+/bug_issues_digest.md
+
+# L-S1: secret-bearing artifacts — block by extension/name so they can never be
+# accidentally committed (TLS material, PKCS#12 bundles, GCP service-account
+# JSON keys). Real values live in Vault / k8s secrets, not the repo.
+*.pem
+*.key
+*.p12
+service-account*.json
+gcs-sa*.json
+
+# Claude Code agent isolation worktrees — ephemeral, never tracked.
+.claude/worktrees
+
+# CI test fixture output — auto-generated, must not be committed
+portal-results.json
+credentials/
+.venv312/

kurious-0.8.4/CHANGELOG.md

@@ -0,0 +1,323 @@
+# Changelog
+
+All notable changes to the `kurious` Python SDK are documented here.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and
+this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.8.4] — 2026-06-12
+
+### Fixed
+- **`projects.ingest(wait=True)` now blocks until the file is searchable
+  (#800)**: a "completed" auto-ingest job only meant preprocessing had
+  finished — the chained indexing job was still running, so an immediate
+  search hit an empty index and returned 0 sources. `wait=True` now follows
+  the chained job recorded in `job.result["auto_chain"]` to a terminal state
+  and then polls the file's unified ingest status until it is `indexed` (or
+  `failed`). The whole wait shares the one `timeout_s` budget.
+- **Real indexing counts on `auto_chain` (#801)**: after the chained job
+  finishes, `job.result["auto_chain"]` carries its terminal `status` plus
+  `documents_indexed` / `documents_failed` from the indexing run (and
+  `error` on failure), so callers can tell whether indexing actually
+  succeeded without scraping worker logs.
+
+## [0.7.0] — 2026-06-04
+
+### Added
+- **`kurious init` CLI (#709)**: a headless onboarding wizard. Prompts for
+  email, verifies it via email OTP, then signs up, logs in, mints a
+  `read_write` API key, and saves it to `~/.kurious/config.toml` (override with
+  `--config` or `KURIOUS_CONFIG_PATH`). No browser required — the entry point
+  for Builder Mode. Installed as the `kurious` console script
+  (`pip install kurious`). Non-interactive: `KURIOUS_PASSWORD` +
+  `KURIOUS_OTP_CODE`. `--skip-otp` skips the client-side OTP prompt only —
+  signup still fails if the server requires verification (not a bypass).
+- **Email OTP methods (#709)**: `auth.request_email_otp(email)` →
+  `EmailOtpRequestResult` and `auth.verify_email_otp(email, code)` →
+  `EmailOtpVerifyResult`. Verify-first signup: prove email ownership, then pass
+  the returned `verification_token` to `signup(..., email_verification_token=)`.
+- **`AIntropy.from_config()`**: build a client from the credentials written by
+  `kurious init`, so scripts can do `client = AIntropy.from_config()`.
+- **`SignupResponse`, `EmailOtpRequestResult`, `EmailOtpVerifyResult` types.**
+
+### Changed (breaking)
+- **`client.auth.signup()` no longer returns tokens (#709)**: signup is async
+  (HTTP 202) — the engine creates the account + org and queues
+  entitlement provisioning, so it returns a `SignupResponse` (`user_id`,
+  `company_id`) instead of a `TokenResponse`, and no longer sets auth state.
+  Call `client.auth.login()` afterwards to obtain a session. This matches the
+  engine fix where `POST /auth/signup` returns the upstream 202 body instead of
+  raising a `ValidationError` trying to coerce it into a token response.
+
+## [0.6.0] — 2026-06-03
+
+### Changed (breaking)
+- **`client.video.inspect()` now requires `project_id` (#657)**: method signature changed from `inspect()` → `inspect(project_id: str)`. Update call sites to pass the project UUID.
+
+### Added
+- `VideoInspectResult.empty: bool` — engine returns HTTP 200 with `empty=True` when the project has no parquet shards yet.
+
+## [0.5.5] — 2026-05-26
+
+### Fixed
+- **`client.video.preprocess()` missing `project_id` (#633)**: the method
+  signature now accepts `project_id` as the first required argument and
+  passes it on the POST body. Earlier versions hit the engine schema
+  with `project_id` missing and got HTTP 422 `ValidationError`. Customers
+  following the LVBench / Video-MME runbook against SDK ≤ 0.5.4 saw the
+  call fail immediately; on 0.5.5 the documented happy path works
+  end-to-end. Paired with an engine-side fix that wires
+  `ensure_project_config_for_ingest` into `POST /pipeline/ingest` so the
+  auto-routing path is searchable as soon as the chain finishes (#634).
+
+### Added
+- `client.projects.ingest()` is now the recommended one-call upload +
+  auto-route + dispatch path. The previous `presign_upload → PUT → files.ingest
+  → files.wait_for_job` flow remains available but `client.files.ingest`,
+  `client.files.wait_for_job` emit `DeprecationWarning`.
+- `client.resources` property — discovery aid that lists every attached
+  resource attribute. `repr(client)` now advertises the resource count.
+- Page-wrapper return types for previously-raw `List[...]` endpoints:
+  `EvalSetPage`, `EvalRunPage`, `EvalResultPage`, `APIKeyPage`,
+  `AuditLogPage`, `IngestionJobPage`. All wrappers are iterable, support
+  `len()` / indexing, and are falsy when empty so existing `for x in page:`
+  call sites keep working.
+- `JobStatus` Literal exported (`"pending" | "running" | "completed" |
+  "failed" | "cancelled"`) for IDE autocomplete on status comparisons.
+- `JobThrottled` and `TERMINAL_STATUSES` exported from the package root
+  (previously importable only via `kurious.jobs`).
+
+### Fixed
+- **Cross-tenant search leak (#592 Layer 4)**: `project_rag_search` no longer
+  falls back to the NJ open data global index when a project-scoped caller
+  passes a `project_config` without `search_mode="kg_unstructured"` —
+  returns empty hits with a clear error instead of another tenant's PDFs.
+- **GCS upload silent failure (#592 Layer 3)**: `client.files.upload` and
+  `client.projects.ingest` now raise `AIntropyError` on a non-2xx PUT
+  response from GCS. Previously a 403 / 413 / 415 was silently absorbed
+  and the downstream ingest "succeeded" on a missing object.
+- **SSE error events**: an `event: error` frame mid-stream now raises
+  `AIntropyError` instead of being yielded as a regular event and dropped
+  by text-only consumers. No more silent stream truncation.
+- **`Job.retry` 429 handling**: the prior `httpx.HTTPStatusError` catch
+  was dead code (the transport raises `RateLimitError` instead). Now
+  correctly translates rate-limit responses into `JobThrottled` with
+  the server's `Retry-After` hint.
+- **JWT auto-refresh**: `client.AIntropy` now wires
+  `auth.ensure_valid_token` into the transport as a pre-request hook,
+  matching the docstring claim. Long-lived JWT-mode clients no longer
+  hit random 401s when the access token expires.
+- **`project_files.status` state machine (#592 Layer 3)**: the engine
+  now transitions `pending_upload → ingesting` at `auto_ingest` dispatch
+  and `ingesting → indexed` at the terminal worker handler. SDK
+  pollers see live state instead of an indefinite `pending_upload`.
+- Worker `video.preprocess` now verifies `assembled.parquet` exists in
+  GCS before declaring success (closes the 121s mark-completed-with-2-of-5-
+  stages regression).
+- Worker `video.preprocess` → `video.ingest` auto-chain wired so a
+  single SDK call ends with the file actually indexed and searchable. The
+  pre-existing `video.ingest → kg.run` chain now propagates `file_id`
+  through to the terminal so `project_files.status=INDEXED` lands.
+- New worker handler `document.preprocess` registered (was a stub) so
+  document uploads via auto-ingest run end-to-end like video.
+
+### Changed
+- `DEFAULT_BASE_URL` docstring corrected: confirms it points at the dev
+  beta host and documents that prod opens at GA.
+- All `wait_for_job` helpers (files, evals, video, nl2sql, kg) now share
+  the canonical `TERMINAL_JOB_STATUSES` frozenset from `kurious.types`.
+  Previously three different definitions disagreed on whether `cancelled`
+  was terminal.
+- README rewritten: `client.whoami()` first-call section, tenant-isolation
+  explainer, `client.projects.ingest()` quickstart replaces the old
+  presign + requests.put flow, ingest-status / resume recovery example,
+  exception table mapping HTTP codes to typed errors, `mode="deep_think"`
+  example, discoverability section (`client.resources`).
+
+## [0.5.4] — 2026-05-21
+
+### Added
+- `Job` now carries auto-ingest detection metadata when the job was created
+  via `POST /pipeline/ingest`: `detected_domain`, `detected_sub_kind`,
+  `detected_mime`, `detected_confidence`, `fallback_used`, `detection_notes`.
+  Previously these fields were silently discarded by `Job.from_payload`;
+  callers can now inspect what the magic-byte detector decided without a
+  separate status call.
+- `AutoIngestConfig` and `KGAutoChainConfig` types exported from the package
+  root. Pass `AutoIngestConfig(kg=KGAutoChainConfig(enabled=False))` to
+  `client.projects.ingest(config=...)` to suppress the automatic KG rebuild
+  after document ingestion. Plain `dict` config still accepted for backward
+  compatibility.
+- `client.files.wait_for_indexed(file_id, *, poll_interval=5.0, timeout=7200.0)`
+  — convenience poller that calls `GET /ingest/status/{file_id}` until
+  `overall_status` is `indexed` or `failed`. Works across both the document
+  chain (`ingestion_jobs`) and video / KG path (`pipeline_jobs`). Raises
+  `TimeoutError` when the deadline elapses.
+- `ProjectFile` now carries optional detection fields (`detected_domain`,
+  `detected_sub_kind`, `detection_confidence`, `detection_notes`) stamped
+  by the auto-routing ingest endpoint. `None` on files uploaded via legacy
+  per-domain endpoints.
+
+## [0.5.3] — 2026-05-19
+
+### Fixed
+- `search.nl2sql`, `search.intelligent`, `search.intelligent_stream`, and
+  `search.intelligent_stream_text` were sending `project_id` only in the
+  URL path. The backend `NL2SQLQueryRequest` / `IntelligentQueryRequest`
+  schemas also require `project_id` in the body, so every call to these
+  four methods returned `422 ValidationError: {type: missing, loc:
+  [body, project_id]}`. The SDK now injects `project_id` into the body
+  payload automatically — caller signatures are unchanged.
+- Synced stale `__version__` in `src/kurious/__init__.py` (was reporting
+  `0.5.1` while `pyproject.toml` advertised `0.5.2`).
+
+## [0.5.2] — 2026-05-15
+
+### Added
+- `client.whoami()` — returns `WhoAmIResponse` with `user_id`, `company_id`,
+  `access_type`, and `project_id`. Backed by `GET /api/v1/whoami`.
+- `client.company_id` property — inspect the resolved company after init.
+- Auto-resolution of `company_id` on `AIntropy(api_key=...)` construction:
+  the client calls `/whoami` eagerly, caches the result on the transport, and
+  injects `X-Company-ID` automatically on every subsequent request. Callers
+  no longer need to pass `company_id` explicitly when using API-key auth.
+
+### Changed
+- `DEFAULT_BASE_URL` corrected to the dev beta host
+  `kurious-backend-dev-api.centralus.cloudapp.azure.com/api/v1` (was the
+  unreachable `api.aintropy.ai`). Prod opens at GA; override `base_url=`
+  on the client when prod is live.
+- `wait_for_job` / `Job.wait_until_done` timeout raised `300s → 7200s` and
+  poll interval increased `2s → 5s` — matches real KG and video pipeline runtimes.
+
+## [0.5.1] — 2026-05-12
+
+### Added
+- `client.projects.ingest(project_id, file_or_uri, ...)` — one-click
+  upload + auto-route + dispatch. Accepts a local path, a `gs://` URI,
+  or a file-like object. Returns a `Job` keyed off `pipeline_jobs.id`.
+  Passing `wait=True` blocks until terminal status and invokes an
+  optional `on_progress` callback on each poll tick.
+- `client.projects.jobs(project_id, ...)` — cursor-paginated iterator
+  over all `Job`s for a project.
+- `client.jobs` — new `JobsResource` with `get(job_id)` (single fetch)
+  and `list(...)` (cross-project paginated iterator).
+- `Job` dataclass exported from the package root with `refresh()`,
+  `retry(publish=False)`, `cancel()`, and `wait_until_done(...)`
+  ergonomics over `GET/POST /api/v1/jobs/{id}[/retry|/cancel]`.
+
+### Notes
+- Backed by `POST /api/v1/pipeline/ingest` (auto-route by magic bytes)
+  and `GET /api/v1/jobs/{id}` (sanitised `error_detail` for non-admins).
+- `retry` / `cancel` are no-ops against an engine that has not been upgraded.
+
+## [0.5.0] — 2026-05-07
+
+### Added
+- `client.projects.get_step_timings(project_id, kind=None)` — per-project
+  per-pipeline-step duration rollup (count / avg_ms / p50 / p95 / total_ms).
+  Optional `kind` parameter filters to a single pipeline kind
+  (e.g. `video.preprocess`, `kg.extract`).
+- New types `StepTiming` and `StepTimingsResponse` exported from the
+  package root.
+
+### Notes
+- Backed by `GET /api/v1/projects/{project_id}/step-timings`. The endpoint
+  is scoped to project members; non-members get 403.
+- Use `client.projects.get_step_timings(project_id).steps` to drive a
+  "where is time going" panel; pair with `total_duration_ms` to compute
+  share-of-time per step.
+
+## [0.4.1] — 2026-05-06
+
+### Added
+- `client.video.backfill_utterances(project_id, max_slices=4, ...)` — kick off
+  the new sliced-scroll utterance backfill (engine endpoint
+  `POST /video/pipeline/backfill_utterances` introduced in PR #409). Returns
+  a `BackfillFanoutResponse` with the umbrella `job_id` and the dispatched
+  `slice_count`. Children run in parallel on the engine's background workers;
+  poll the umbrella with `client.video.get_job(job_id)`.
+- `BackfillFanoutResponse` and `JobChildrenRollup` types exported from the
+  package root.
+- `VideoPipelineStatus.children` convenience property — surfaces
+  `{slice_count, completed, failed, running, pending}` when the polled row is
+  an umbrella over fan-out children. Returns `None` for single-task jobs.
+- `VideoPipelineJob` now carries dispatch metadata: `celery_task_id`,
+  `priority`, `queue_name`, `parent_task_id`, `slice_id`, `max_slices`.
+  All optional — only populated on rows created after the engine enables
+  async dispatch.
+
+### Notes
+- `max_slices` is capped at 16 by the engine; the SDK validates client-side
+  and raises `ValueError` before any HTTP call.
+
+## [0.4.0] — 2026-05-04
+
+### Added
+- `client.video.preprocess(input_video_gcs_uri, stage_prefix, ...)` — kick off
+  the per-video preprocessing chain (decode → transcribe → embed → caption →
+  assemble). Per-stage `skip_*` flags allow resuming after an interruption.
+- New endpoint backing it: `POST /video/pipeline/preprocess`.
+
+### Notes
+- Per-video runtime varies based on video length and available GPU resources.
+- Intermediate parquets land at
+  `<stage_prefix>/<video_id>/{frames,transcripts,embeddings,captions,assembled}.parquet`.
+  Downstream `client.video.ingest()` consumes the assembled parquet.
+
+## [0.3.0] — 2026-05-01
+
+### Added
+- `client.video.timeline(video_id, max_events=100)` — fetch per-video
+  scene-level timeline (chunked events with start/end timestamps, topic,
+  keywords, speakers). Powers timeline hover-cards on video result UIs.
+- `client.video.cancel_job(job_id)` — cancel a pending or running video
+  pipeline job.
+- New types: `VideoTimeline`, `VideoTimelineEvent`.
+
+### Coverage
+SDK now covers ~98% of the public Kurious Engine API surface. The
+`/projects/{id}/ingest*` upload + ingestion endpoints (already in
+`client.files.*` since 0.1.0) remain the recommended path for uploads;
+no additional ingestion wrapper is needed.
+
+## [0.2.0] — 2026-04-25
+
+### Added
+- `client.conversations` — chat thread management: create / list / get / update /
+  delete, message CRUD, public-share enable/disable, public shared-conversation
+  reads.
+- `client.bookmarks` — list / create / delete saved messages.
+- `client.search_log` — log a UI search and submit star-rating feedback.
+- `client.ai` — direct embedding and chat-completion access.
+- `client.entitlements` — list, get, request upgrade, start trial.
+- `search.blended_query()` — benchmark blended (lexical + semantic) search.
+- `search.nl2sql_stream()` — streaming NL2SQL responses.
+- `py.typed` marker so type checkers (mypy, pyright) recognize the package as
+  fully typed.
+- `__repr__` on every resource class for friendlier debugging.
+- `LICENSE` (MIT) and richer `pyproject.toml` metadata: project URLs, keywords,
+  `lint` / `docs` extras groups, ruff and mypy config.
+
+### Changed
+- Expanded `README.md` with end-to-end examples (auth, projects, ingest,
+  search, conversations, evals, error handling, streaming).
+- Tightened class docstrings on `IndexStats` and other low-level types.
+
+### Coverage
+SDK now covers ~95% of the public Kurious Engine API surface (vs. 71% in
+v0.1.0). Admin- and internal-only endpoints (`/admin/*`, `/internal/*`,
+`/telemetry/*`, `/resources/*`) remain intentionally out of scope.
+
+## [0.1.0] — 2026-04-22
+
+Initial release.
+
+### Added
+- Core resources: `auth`, `projects`, `files`, `indices`, `documents`, `search`,
+  `evals`, `usage`, `api_keys`, `nl2sql`, `kg`, `video`.
+- Sync HTTP transport with retries, rate-limit backoff, and JWT/API-key auth.
+- SSE streaming helpers for `intelligent_stream` / `intelligent_stream_text`.
+- Pydantic v2 type models mirroring Kurious Engine response shapes.

kurious-0.8.4/LICENSE

@@ -0,0 +1,6 @@
+Copyright © 2026 AIntropy. All rights reserved.
+
+This software and its source code are proprietary and confidential.
+Unauthorized use, copying, modification, distribution, or redistribution
+of this software, in whole or in part, is strictly prohibited without the
+prior written consent of AIntropy.