contextgem 0.25.0__tar.gz → 0.26.0__tar.gz

{contextgem-0.25.0 → contextgem-0.26.0}/.gitignore

@@ -24,3 +24,6 @@ dist
 code_sample.py
 
 settings.local.json
+
+.claude
+.agents

contextgem-0.26.0/LICENSES.md

@@ -0,0 +1,115 @@
+# License Analysis Report
+
+**Project license:** Apache-2.0
+
+|Package|Ecosystem|Group|Source|License|Risk|Status|
+|---------|-----------|-------|--------|---------|------|--------|
+|[aiolimiter (1.2.1)](https://pypi.org/project/aiolimiter/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[colorlog (6.12.0)](https://pypi.org/project/colorlog/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [colorama (0.4.6)](https://pypi.org/project/colorama/) (also: litellm, openai +1 more)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[docstring-parser (0.18.0)](https://pypi.org/project/docstring-parser/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[fastjsonschema (2.22.1)](https://pypi.org/project/fastjsonschema/)|python|prod|[pyproject.toml](pyproject.toml)|[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[genai-prices (0.0.72)](https://pypi.org/project/genai-prices/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [annotated-types (0.8.0)](https://pypi.org/project/annotated-types/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [anyio (4.14.2)](https://pypi.org/project/anyio/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [exceptiongroup (1.3.1)](https://pypi.org/project/exceptiongroup/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [h11 (0.16.0)](https://pypi.org/project/h11/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [httpcore2 (2.9.1)](https://pypi.org/project/httpcore2/)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [httpx2 (2.9.1)](https://pypi.org/project/httpx2/)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [idna (3.18.0)](https://pypi.org/project/idna/) (also: litellm, openai +1 more)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [pydantic-core (2.46.4)](https://pypi.org/project/pydantic-core/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [truststore (0.10.4)](https://pypi.org/project/truststore/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [typing-inspection (0.4.2)](https://pypi.org/project/typing-inspection/) (also: litellm, openai +1 more)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[jinja2 (3.1.6)](https://pypi.org/project/jinja2/)|python|prod|[pyproject.toml](pyproject.toml)|[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [markupsafe (3.0.3)](https://pypi.org/project/markupsafe/) (also: litellm)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[litellm (1.93.0)](https://pypi.org/project/litellm/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [aiohappyeyeballs (2.7.1)](https://pypi.org/project/aiohappyeyeballs/)|python|prod||[PSF-2.0](https://spdx.org/licenses/PSF-2.0.html)|permissive|✓|
+|  └─ [aiohttp (3.14.3)](https://pypi.org/project/aiohttp/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html) AND [MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [aiosignal (1.4.0)](https://pypi.org/project/aiosignal/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [async-timeout (5.0.1)](https://pypi.org/project/async-timeout/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [attrs (26.1.0)](https://pypi.org/project/attrs/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [certifi (2026.7.22)](https://pypi.org/project/certifi/) (also: openai, wtpsplit-lite)|python|prod||[MPL-2.0](https://spdx.org/licenses/MPL-2.0.html)|weak-copyleft|⚠|
+|  └─ [charset-normalizer (3.4.9)](https://pypi.org/project/charset-normalizer/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [click (8.4.2)](https://pypi.org/project/click/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [distro (1.9.0)](https://pypi.org/project/distro/) (also: openai)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [fastuuid (0.14.0)](https://pypi.org/project/fastuuid/)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [filelock (3.32.0)](https://pypi.org/project/filelock/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [frozenlist (1.8.0)](https://pypi.org/project/frozenlist/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [fsspec (2026.7.0)](https://pypi.org/project/fsspec/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [hf-xet (1.5.2)](https://pypi.org/project/hf-xet/) (also: wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [httpcore (1.0.9)](https://pypi.org/project/httpcore/) (also: openai, wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [httpx (0.28.1)](https://pypi.org/project/httpx/) (also: openai, wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [huggingface-hub (1.25.1)](https://pypi.org/project/huggingface-hub/) (also: wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [importlib-metadata (8.9.0)](https://pypi.org/project/importlib-metadata/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [jiter (0.16.0)](https://pypi.org/project/jiter/) (also: openai)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [jsonschema (4.26.0)](https://pypi.org/project/jsonschema/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [jsonschema-specifications (2025.9.1)](https://pypi.org/project/jsonschema-specifications/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [multidict (6.7.1)](https://pypi.org/project/multidict/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [packaging (26.2.0)](https://pypi.org/project/packaging/) (also: onnxruntime, wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html) OR [BSD-2-Clause](https://spdx.org/licenses/BSD-2-Clause.html)|permissive|✓|
+|  └─ [propcache (0.5.2)](https://pypi.org/project/propcache/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [pyyaml (6.0.3)](https://pypi.org/project/pyyaml/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [referencing (0.37.0)](https://pypi.org/project/referencing/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [regex (2026.7.19)](https://pypi.org/project/regex/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html) AND [CNRI-Python](https://spdx.org/licenses/CNRI-Python.html)|unknown|?|
+|  └─ [requests (2.34.2)](https://pypi.org/project/requests/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [rpds-py (0.30.0)](https://pypi.org/project/rpds-py/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [rpds-py (2026.6.3)](https://pypi.org/project/rpds-py/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [sniffio (1.3.1)](https://pypi.org/project/sniffio/) (also: openai)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html) OR [MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [tiktoken (0.13.0)](https://pypi.org/project/tiktoken/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [tokenizers (0.23.1)](https://pypi.org/project/tokenizers/) (also: wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [tqdm (4.70.0)](https://pypi.org/project/tqdm/) (also: openai, wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html) AND [MPL-2.0](https://spdx.org/licenses/MPL-2.0.html)|weak-copyleft|⚠|
+|  └─ [urllib3 (2.7.0)](https://pypi.org/project/urllib3/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [yarl (1.24.5)](https://pypi.org/project/yarl/)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [zipp (4.1.0)](https://pypi.org/project/zipp/)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[lxml (6.1.1)](https://pypi.org/project/lxml/)|python|prod|[pyproject.toml](pyproject.toml)|[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[onnxruntime (1.23.2)](https://pypi.org/project/onnxruntime/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [coloredlogs (15.0.1)](https://pypi.org/project/coloredlogs/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [flatbuffers (25.12.19)](https://pypi.org/project/flatbuffers/) (also: wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [humanfriendly (10.0.0)](https://pypi.org/project/humanfriendly/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [mpmath (1.3.0)](https://pypi.org/project/mpmath/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [numpy (2.2.6)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [numpy (2.4.6)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[0BSD](https://spdx.org/licenses/0BSD.html) AND [BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html) AND [CC0-1.0](https://spdx.org/licenses/CC0-1.0.html) AND [MIT](https://spdx.org/licenses/MIT.html) AND [Zlib](https://spdx.org/licenses/Zlib.html)|permissive|✓|
+|  └─ [numpy (2.5.1)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[0BSD](https://spdx.org/licenses/0BSD.html) AND [BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html) AND [CC0-1.0](https://spdx.org/licenses/CC0-1.0.html) AND [MIT](https://spdx.org/licenses/MIT.html) AND [Zlib](https://spdx.org/licenses/Zlib.html)|permissive|✓|
+|  └─ [protobuf (7.35.1)](https://pypi.org/project/protobuf/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [pyreadline3 (3.5.6)](https://pypi.org/project/pyreadline3/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [sympy (1.14.0)](https://pypi.org/project/sympy/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[onnxruntime (1.28.0)](https://pypi.org/project/onnxruntime/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [coloredlogs (15.0.1)](https://pypi.org/project/coloredlogs/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [flatbuffers (25.12.19)](https://pypi.org/project/flatbuffers/) (also: wtpsplit-lite)|python|prod||[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|  └─ [humanfriendly (10.0.0)](https://pypi.org/project/humanfriendly/) (also: wtpsplit-lite)|python|prod||[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|  └─ [mpmath (1.3.0)](https://pypi.org/project/mpmath/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [numpy (2.2.6)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [numpy (2.4.6)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[0BSD](https://spdx.org/licenses/0BSD.html) AND [BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html) AND [CC0-1.0](https://spdx.org/licenses/CC0-1.0.html) AND [MIT](https://spdx.org/licenses/MIT.html) AND [Zlib](https://spdx.org/licenses/Zlib.html)|permissive|✓|
+|  └─ [numpy (2.5.1)](https://pypi.org/project/numpy/) (also: wtpsplit-lite)|python|prod||[0BSD](https://spdx.org/licenses/0BSD.html) AND [BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html) AND [CC0-1.0](https://spdx.org/licenses/CC0-1.0.html) AND [MIT](https://spdx.org/licenses/MIT.html) AND [Zlib](https://spdx.org/licenses/Zlib.html)|permissive|✓|
+|  └─ [protobuf (7.35.1)](https://pypi.org/project/protobuf/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [pyreadline3 (3.5.6)](https://pypi.org/project/pyreadline3/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|  └─ [sympy (1.14.0)](https://pypi.org/project/sympy/) (also: wtpsplit-lite)|python|prod||[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[openai (2.49.0)](https://pypi.org/project/openai/)|python|prod|[pyproject.toml](pyproject.toml)|[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|[pillow (11.3.0)](https://pypi.org/project/pillow/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT-CMU](https://spdx.org/licenses/MIT-CMU.html)|permissive|✓|
+|[pydantic (2.13.4)](https://pypi.org/project/pydantic/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[python-dotenv (1.2.2)](https://pypi.org/project/python-dotenv/)|python|prod|[pyproject.toml](pyproject.toml)|[BSD-3-Clause](https://spdx.org/licenses/BSD-3-Clause.html)|permissive|✓|
+|[python-ulid (3.2.1)](https://pypi.org/project/python-ulid/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[tenacity (9.1.4)](https://pypi.org/project/tenacity/)|python|prod|[pyproject.toml](pyproject.toml)|[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)|permissive|✓|
+|[tethered (0.5.1)](https://pypi.org/project/tethered/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+|[typing-extensions (4.16.0)](https://pypi.org/project/typing-extensions/)|python|prod|[pyproject.toml](pyproject.toml)|[PSF-2.0](https://spdx.org/licenses/PSF-2.0.html)|permissive|✓|
+|[wtpsplit-lite (0.2.0)](https://pypi.org/project/wtpsplit-lite/)|python|prod|[pyproject.toml](pyproject.toml)|[MIT](https://spdx.org/licenses/MIT.html)|permissive|✓|
+
+**Summary:** 0 violations, 2 warnings, 1 unknown, 75 ok (of which 4 reviewed)
+
+**Completed in:** 0.66s
+
+## Details
+
+- ⚠ certifi uses MPL-2.0 — weak copyleft license — modifications to the licensed files must remain under the same license; linking from other files is allowed. Review whether this is acceptable for your Apache-2.0 project
+- ? regex uses Apache-2.0 AND CNRI-Python — could not determine compatibility with Apache-2.0
+- ⚠ tqdm uses MIT AND MPL-2.0 — weak copyleft license — modifications to the licensed files must remain under the same license; linking from other files is allowed. Review whether this is acceptable for your Apache-2.0 project
+
+## Reviewed
+
+|Package|Detected|Reviewed|Note|
+|---------|----------|----------|----|
+|[certifi (2026.7.22)](https://pypi.org/project/certifi/)|MPL-2.0|[MPL-2.0](https://spdx.org/licenses/MPL-2.0.html)|Apache-2.0 library distributed via PyPI; certifi is an unmodified upstream dependency used with dynamic linkage. MPL-2.0 is file-level weak copyleft that only triggers on modification of the MPL-licensed files — no modifications, so the obligation is dormant.|
+|[regex (2026.7.19)](https://pypi.org/project/regex/)|Apache-2.0 AND CNRI-Python|[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html) AND [CNRI-Python](https://spdx.org/licenses/CNRI-Python.html)|Flagged UNKNOWN only because licenseal cannot classify the compound expression. CNRI-Python is an OSI-approved permissive license (the historical Python license) and is compatible with Apache-2.0; regex is used unmodified with dynamic linkage. Detection-classification override.|
+|[tqdm (4.70.0)](https://pypi.org/project/tqdm/)|MIT AND MPL-2.0|[MIT](https://spdx.org/licenses/MIT.html) AND [MPL-2.0](https://spdx.org/licenses/MPL-2.0.html)|Same rationale as certifi: unmodified upstream dependency, dynamic linkage. The MPL-2.0 portion is file-level weak copyleft and does not trigger absent modification of the MPL-licensed files.|
+|[wtpsplit-lite (0.2.0)](https://pypi.org/project/wtpsplit-lite/)|UNKNOWN|[MIT](https://spdx.org/licenses/MIT.html)|PyPI package ships no license metadata (hence UNKNOWN); the repository LICENSE file is MIT, confirmed via GitHub's license-detection API for the superlinear-ai/wtpsplit-lite repo (spdx_id MIT). Detection override — MIT is permissive and compatible with Apache-2.0.|
+
+_Generated by `licenseal check`. Versions shown are the resolved releases checked._

{contextgem-0.25.0 → contextgem-0.26.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: contextgem
-Version: 0.25.0
+Version: 0.26.0
 Summary: Effortless LLM extraction from documents
 Project-URL: Homepage, https://github.com/shcherbak-ai/contextgem
 Project-URL: Documentation, https://github.com/shcherbak-ai/contextgem/blob/main/README.md
@@ -10,6 +10,7 @@ Author-email: shcherbak-ai <sergii@shcherbak.ai>
 Maintainer-email: shcherbak-ai <sergii@shcherbak.ai>
 License: Apache-2.0
 License-File: LICENSE
+License-File: LICENSES.md
 License-File: NOTICE
 Keywords: artificial-intelligence,aspect-extraction,automated-prompting,concept-extraction,content-extraction,context-aware,contextgem,contract-analysis,contract-automation,contract-intelligence,contract-management,contract-parsing,contract-review,data-extraction,document,document-analysis,document-extraction,document-intelligence,document-parsing,document-pipeline,document-processing,document-qa,document-understanding,entity-extraction,extraction-justifications,extraction-pipeline,fintech,generative-ai,information-extraction,insights-extraction,knowledge-extraction,large-language-models,legaltech,llm,llm-extraction,llm-framework,llm-library,llm-reasoning,low-code,machine-learning,multilingual,multimodal,neural-segmentation,nlp,no-prompt-engineering,prompt-free,question-answering,reference-mapping,semantic-analysis,structured-data,structured-data-extraction,text-analysis,text-processing,topic-extraction,unstructured-data,zero-shot
 Classifier: Development Status :: 4 - Beta
@@ -41,13 +42,13 @@ Requires-Dist: aiolimiter<2.0.0,>=1.2.1
 Requires-Dist: colorlog<7.0.0,>=6.9.0
 Requires-Dist: docstring-parser<1.0.0,>=0.17.0
 Requires-Dist: fastjsonschema<3.0.0,>=2.21.2
-Requires-Dist: genai-prices==0.0.59
+Requires-Dist: genai-prices==0.0.72
 Requires-Dist: jinja2<4.0.0,>=3.1.5
-Requires-Dist: litellm==1.83.14
+Requires-Dist: litellm==1.93.0
 Requires-Dist: lxml<7.0.0,>=5.4.0
 Requires-Dist: onnxruntime<1.24.0; python_version < '3.11'
 Requires-Dist: onnxruntime>=1.24.0; python_version >= '3.14'
-Requires-Dist: openai==2.24.0
+Requires-Dist: openai==2.49.0
 Requires-Dist: pillow<12.0.0,>=11.3.0
 Requires-Dist: pydantic<3.0.0,>=2.10.6
 Requires-Dist: python-ulid<4.0.0,>=3.0.0
@@ -64,7 +65,7 @@ Description-Content-Type: text/markdown
 |          |        |
 |----------|--------|
 | **Package** | [![PyPI](https://img.shields.io/pypi/v/contextgem?logo=pypi&label=PyPi&logoColor=gold&style=flat-square)](https://pypi.org/project/contextgem/) [![PyPI Downloads](https://static.pepy.tech/badge/contextgem/month)](https://pepy.tech/projects/contextgem) [![Python Versions](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13%20%7C%203.14-blue?logo=python&logoColor=gold&style=flat-square)](https://www.python.org/downloads/) [![License](https://img.shields.io/badge/License-Apache_2.0-blue?style=flat-square)](https://opensource.org/licenses/Apache-2.0) |
-| **Quality** | [![tests](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/ci-tests.yml?branch=main&style=flat-square&label=tests)](https://github.com/shcherbak-ai/contextgem/actions/workflows/ci-tests.yml) [![Coverage](https://img.shields.io/endpoint?url=https://gist.githubusercontent.com/SergiiShcherbak/daaee00e1dfff7a29ca10a922ec3becd/raw/coverage.json&style=flat-square)](https://github.com/shcherbak-ai/contextgem/actions) [![CodeQL](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/codeql.yml?branch=main&style=flat-square&label=CodeQL)](https://github.com/shcherbak-ai/contextgem/actions/workflows/codeql.yml) [![security: bandit](https://img.shields.io/badge/security-bandit-yellow.svg?style=flat-square)](https://github.com/PyCQA/bandit) [![OpenSSF Best Practices](https://www.bestpractices.dev/projects/10489/badge?1)](https://www.bestpractices.dev/projects/10489) |
+| **Quality** | [![tests](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/ci-tests.yml?branch=main&style=flat-square&label=tests)](https://github.com/shcherbak-ai/contextgem/actions/workflows/ci-tests.yml) [![Coverage](https://img.shields.io/endpoint?url=https://gist.githubusercontent.com/SergiiShcherbak/daaee00e1dfff7a29ca10a922ec3becd/raw/coverage.json&style=flat-square)](https://github.com/shcherbak-ai/contextgem/actions) [![CodeQL](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/codeql.yml?branch=main&style=flat-square&label=CodeQL)](https://github.com/shcherbak-ai/contextgem/actions/workflows/codeql.yml) [![license compatibility](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/licenseal.yml?branch=main&style=flat-square&label=licenses)](https://github.com/shcherbak-ai/contextgem/actions/workflows/licenseal.yml) [![security: bandit](https://img.shields.io/badge/security-bandit-yellow.svg?style=flat-square)](https://github.com/PyCQA/bandit) [![OpenSSF Best Practices](https://www.bestpractices.dev/projects/10489/badge?1)](https://www.bestpractices.dev/projects/10489) |
 | **Tools** | [![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json&style=flat-square)](https://github.com/astral-sh/uv) [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json&style=flat-square)](https://github.com/astral-sh/ruff) [![Pydantic v2](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/pydantic/pydantic/main/docs/badge/v2.json&style=flat-square)](https://pydantic.dev) [![ty](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ty/main/assets/badge/v0.json&style=flat-square)](https://github.com/astral-sh/ty) [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-blue?logo=pre-commit&logoColor=white&style=flat-square)](https://github.com/pre-commit/pre-commit) [![deptry](https://img.shields.io/badge/deptry-checked-blue?style=flat-square)](https://github.com/fpgmaas/deptry) [![egress: tethered](https://img.shields.io/badge/egress-tethered-orange?labelColor=4B8BBE)](https://github.com/shcherbak-ai/tethered) [![Hatch project](https://img.shields.io/badge/%F0%9F%A5%9A-Hatch-4051b5.svg?style=flat-square)](https://github.com/pypa/hatch) |
 | **Docs** | [![docs](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/docs.yml?branch=main&style=flat-square&label=docs)](https://github.com/shcherbak-ai/contextgem/actions/workflows/docs.yml) [![documentation](https://img.shields.io/badge/docs-latest-blue?style=flat-square)](https://shcherbak-ai.github.io/contextgem/) ![Docstring Coverage](https://contextgem.dev/_static/interrogate-badge.svg) [![DeepWiki](https://img.shields.io/static/v1?label=DeepWiki&message=Chat%20with%20Code&labelColor=%23283593&color=%237E57C2&style=flat-square)](https://deepwiki.com/shcherbak-ai/contextgem) |
 | **Community** | [![Contributor Covenant](https://img.shields.io/badge/Contributor%20Covenant-2.1-4baaaa?style=flat-square)](CODE_OF_CONDUCT.md) [![GitHub issues closed](https://img.shields.io/github/issues-closed/shcherbak-ai/contextgem?style=flat-square)](https://github.com/shcherbak-ai/contextgem/issues?q=is%3Aissue+is%3Aclosed) [![GitHub latest commit](https://img.shields.io/github/last-commit/shcherbak-ai/contextgem?label=latest%20commit&style=flat-square)](https://github.com/shcherbak-ai/contextgem/commits/main) |
@@ -421,4 +422,5 @@ Built with ❤️ in Oslo, Norway.
 
 | Package | Description |
 |---------|-------------|
+| **[licenseal](https://github.com/shcherbak-ai/licenseal)** [![PyPI](https://img.shields.io/pypi/v/licenseal?logo=pypi&logoColor=gold&style=flat-square)](https://pypi.org/project/licenseal/) [![License: Apache-2.0](https://img.shields.io/badge/license-Apache--2.0-blue.svg?style=flat-square)](https://github.com/shcherbak-ai/licenseal/blob/main/LICENSE) | Fast cross-ecosystem license compatibility checker for CI, audits, and enterprise adoption. Scans manifests, lockfiles, and public registry metadata without installing dependencies. |
 | **[tethered](https://github.com/shcherbak-ai/tethered)** [![PyPI](https://img.shields.io/pypi/v/tethered?logo=pypi&logoColor=gold&style=flat-square)](https://pypi.org/project/tethered/) [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg?style=flat-square)](https://github.com/shcherbak-ai/tethered/blob/main/LICENSE) | Runtime network egress control for Python. One function call blocks all unauthorized outbound connections — zero dependencies, no infrastructure changes. Ideal for supply chain defense, AI agent guardrails, and test isolation. |

{contextgem-0.25.0 → contextgem-0.26.0}/README.md

@@ -5,7 +5,7 @@
 |          |        |
 |----------|--------|
 | **Package** | [![PyPI](https://img.shields.io/pypi/v/contextgem?logo=pypi&label=PyPi&logoColor=gold&style=flat-square)](https://pypi.org/project/contextgem/) [![PyPI Downloads](https://static.pepy.tech/badge/contextgem/month)](https://pepy.tech/projects/contextgem) [![Python Versions](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13%20%7C%203.14-blue?logo=python&logoColor=gold&style=flat-square)](https://www.python.org/downloads/) [![License](https://img.shields.io/badge/License-Apache_2.0-blue?style=flat-square)](https://opensource.org/licenses/Apache-2.0) |
-| **Quality** | [![tests](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/ci-tests.yml?branch=main&style=flat-square&label=tests)](https://github.com/shcherbak-ai/contextgem/actions/workflows/ci-tests.yml) [![Coverage](https://img.shields.io/endpoint?url=https://gist.githubusercontent.com/SergiiShcherbak/daaee00e1dfff7a29ca10a922ec3becd/raw/coverage.json&style=flat-square)](https://github.com/shcherbak-ai/contextgem/actions) [![CodeQL](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/codeql.yml?branch=main&style=flat-square&label=CodeQL)](https://github.com/shcherbak-ai/contextgem/actions/workflows/codeql.yml) [![security: bandit](https://img.shields.io/badge/security-bandit-yellow.svg?style=flat-square)](https://github.com/PyCQA/bandit) [![OpenSSF Best Practices](https://www.bestpractices.dev/projects/10489/badge?1)](https://www.bestpractices.dev/projects/10489) |
+| **Quality** | [![tests](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/ci-tests.yml?branch=main&style=flat-square&label=tests)](https://github.com/shcherbak-ai/contextgem/actions/workflows/ci-tests.yml) [![Coverage](https://img.shields.io/endpoint?url=https://gist.githubusercontent.com/SergiiShcherbak/daaee00e1dfff7a29ca10a922ec3becd/raw/coverage.json&style=flat-square)](https://github.com/shcherbak-ai/contextgem/actions) [![CodeQL](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/codeql.yml?branch=main&style=flat-square&label=CodeQL)](https://github.com/shcherbak-ai/contextgem/actions/workflows/codeql.yml) [![license compatibility](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/licenseal.yml?branch=main&style=flat-square&label=licenses)](https://github.com/shcherbak-ai/contextgem/actions/workflows/licenseal.yml) [![security: bandit](https://img.shields.io/badge/security-bandit-yellow.svg?style=flat-square)](https://github.com/PyCQA/bandit) [![OpenSSF Best Practices](https://www.bestpractices.dev/projects/10489/badge?1)](https://www.bestpractices.dev/projects/10489) |
 | **Tools** | [![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json&style=flat-square)](https://github.com/astral-sh/uv) [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json&style=flat-square)](https://github.com/astral-sh/ruff) [![Pydantic v2](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/pydantic/pydantic/main/docs/badge/v2.json&style=flat-square)](https://pydantic.dev) [![ty](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ty/main/assets/badge/v0.json&style=flat-square)](https://github.com/astral-sh/ty) [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-blue?logo=pre-commit&logoColor=white&style=flat-square)](https://github.com/pre-commit/pre-commit) [![deptry](https://img.shields.io/badge/deptry-checked-blue?style=flat-square)](https://github.com/fpgmaas/deptry) [![egress: tethered](https://img.shields.io/badge/egress-tethered-orange?labelColor=4B8BBE)](https://github.com/shcherbak-ai/tethered) [![Hatch project](https://img.shields.io/badge/%F0%9F%A5%9A-Hatch-4051b5.svg?style=flat-square)](https://github.com/pypa/hatch) |
 | **Docs** | [![docs](https://img.shields.io/github/actions/workflow/status/shcherbak-ai/contextgem/docs.yml?branch=main&style=flat-square&label=docs)](https://github.com/shcherbak-ai/contextgem/actions/workflows/docs.yml) [![documentation](https://img.shields.io/badge/docs-latest-blue?style=flat-square)](https://shcherbak-ai.github.io/contextgem/) ![Docstring Coverage](https://contextgem.dev/_static/interrogate-badge.svg) [![DeepWiki](https://img.shields.io/static/v1?label=DeepWiki&message=Chat%20with%20Code&labelColor=%23283593&color=%237E57C2&style=flat-square)](https://deepwiki.com/shcherbak-ai/contextgem) |
 | **Community** | [![Contributor Covenant](https://img.shields.io/badge/Contributor%20Covenant-2.1-4baaaa?style=flat-square)](CODE_OF_CONDUCT.md) [![GitHub issues closed](https://img.shields.io/github/issues-closed/shcherbak-ai/contextgem?style=flat-square)](https://github.com/shcherbak-ai/contextgem/issues?q=is%3Aissue+is%3Aclosed) [![GitHub latest commit](https://img.shields.io/github/last-commit/shcherbak-ai/contextgem?label=latest%20commit&style=flat-square)](https://github.com/shcherbak-ai/contextgem/commits/main) |
@@ -362,4 +362,5 @@ Built with ❤️ in Oslo, Norway.
 
 | Package | Description |
 |---------|-------------|
+| **[licenseal](https://github.com/shcherbak-ai/licenseal)** [![PyPI](https://img.shields.io/pypi/v/licenseal?logo=pypi&logoColor=gold&style=flat-square)](https://pypi.org/project/licenseal/) [![License: Apache-2.0](https://img.shields.io/badge/license-Apache--2.0-blue.svg?style=flat-square)](https://github.com/shcherbak-ai/licenseal/blob/main/LICENSE) | Fast cross-ecosystem license compatibility checker for CI, audits, and enterprise adoption. Scans manifests, lockfiles, and public registry metadata without installing dependencies. |
 | **[tethered](https://github.com/shcherbak-ai/tethered)** [![PyPI](https://img.shields.io/pypi/v/tethered?logo=pypi&logoColor=gold&style=flat-square)](https://pypi.org/project/tethered/) [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg?style=flat-square)](https://github.com/shcherbak-ai/tethered/blob/main/LICENSE) | Runtime network egress control for Python. One function call blocks all unauthorized outbound connections — zero dependencies, no infrastructure changes. Ideal for supply chain defense, AI agent guardrails, and test isolation. |

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/__init__.py

@@ -20,7 +20,7 @@
 ContextGem - Effortless LLM extraction from documents
 """
 
-__version__ = "0.25.0"
+__version__ = "0.26.0"
 __author__ = "Shcherbak AI AS"
 
 from contextgem.public import (

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/base/aspects.py

@@ -219,6 +219,28 @@ class _Aspect(
             if aspect.aspects:
                 aspect._validate_and_process_sub_aspects(aspect.aspects)  # ty: ignore[invalid-argument-type]
 
+    def _reset_nesting_levels(self, level: int = 0) -> None:
+        """
+        Recursively (re)assigns nesting levels, treating this aspect as residing
+        at ``level`` and its sub-aspects at the subsequent levels.
+
+        ``_nesting_level`` is assigned relative to the parent when an aspect is
+        attached as a sub-aspect, and is never otherwise reset. When an aspect
+        that was previously used as a sub-aspect (nesting level > 0) is reused as
+        a top-level aspect (e.g. attached to a ``Document``), its stale level
+        must be normalized so that nesting-level validation and sub-aspect
+        extraction behave correctly. Top-level containers call this with
+        ``level=0``.
+
+        :param level: The nesting level to assign to this aspect.
+        :type level: int
+        :return: None
+        :rtype: None
+        """
+        self._nesting_level = level
+        for sub_aspect in self.aspects:
+            sub_aspect._reset_nesting_levels(level + 1)
+
     @model_validator(mode="after")
     def _validate_aspect_post(self) -> Self:
         """

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/base/documents.py

@@ -28,7 +28,13 @@ from copy import deepcopy
 from pathlib import Path
 from typing import Annotated, Any, Literal
 
-from pydantic import BeforeValidator, Field, field_validator, model_validator
+from pydantic import (
+    BeforeValidator,
+    Field,
+    PrivateAttr,
+    field_validator,
+    model_validator,
+)
 from typing_extensions import Self
 
 from contextgem.internal.base.aspects import _Aspect
@@ -111,6 +117,12 @@ class _Document(_AssignedInstancesProcessor, _MarkdownTextAttributesProcessor):
         ),
     )
 
+    # When True, top-level aspect nesting normalization is suppressed. Set only by
+    # the extraction engine on the throwaway single-aspect documents it builds from
+    # existing (already-nested) aspects, where the added aspect must keep its
+    # nesting level so it still compares equal to the instance being processed.
+    _suppress_aspect_nesting_normalization: bool = PrivateAttr(default=False)
+
     def __setattr__(self, name: str, value: Any) -> None:
         """
         Sets the attribute of an instance, with additional restrictions on specific attributes.
@@ -161,6 +173,34 @@ class _Document(_AssignedInstancesProcessor, _MarkdownTextAttributesProcessor):
         self._set_text_from_paras()
         self._segment_document_text()
 
+    def _normalize_top_level_aspect_nesting(self) -> None:
+        """
+        Normalizes the nesting levels of the document's top-level aspects to 0
+        (propagating to sub-aspects).
+
+        Aspects attached to a document are top-level. An aspect previously used
+        as a sub-aspect carries a non-zero ``_nesting_level``; left unchanged,
+        that stale level prevents assigning sub-aspects to it and breaks
+        nesting-level validation during extraction. Runs on construction and on
+        every aspect (re)assignment via the post-validation hook, so all
+        user-facing paths (constructor, ``aspects`` setter, ``add_aspects``,
+        ``assign_pipeline``) are covered.
+
+        Skipped when ``_suppress_aspect_nesting_normalization`` is set, i.e. on
+        the throwaway single-aspect documents the extraction engine builds from
+        already-nested aspects — those add an existing sub-aspect and rely on it
+        comparing equal (including ``_nesting_level``) to the instance being
+        processed.
+
+        :return: None
+        :rtype: None
+        """
+        if self._suppress_aspect_nesting_normalization:
+            return
+        for aspect in self.aspects:
+            if aspect._nesting_level != 0:
+                aspect._reset_nesting_levels(0)
+
     def assign_pipeline(
         self,
         pipeline: _ExtractionPipeline | _DocumentPipeline,
@@ -479,6 +519,28 @@ class _Document(_AssignedInstancesProcessor, _MarkdownTextAttributesProcessor):
         :raises ValueError: If paragraphs are not matched or ordered correctly in
             the document text, or if `_md_text` is provided without `raw_text`.
         """
+        self._run_post_validation()
+        return self
+
+    def _run_post_validation(self) -> None:
+        """
+        Performs document post-validation: normalizes top-level aspect nesting
+        levels and checks paragraph/markdown consistency.
+
+        Implemented as a plain method (separate from the ``model_validator``
+        wrapper) so it can also be invoked during deserialization, once private
+        attributes such as ``_md_text`` have been restored.
+
+        :return: None
+        :rtype: None
+        :raises ValueError: If paragraphs are not matched or ordered correctly in
+            the document text, or if `_md_text` is provided without `raw_text`.
+        """
+
+        # Aspects attached to a document are top-level: normalize their nesting
+        # levels (no-op when already correct, or when suppressed for the
+        # extraction engine's internal single-aspect documents).
+        self._normalize_top_level_aspect_nesting()
 
         # Validate that _md_text cannot be populated without raw_text
         if self._md_text and not self.raw_text:
@@ -516,4 +578,19 @@ class _Document(_AssignedInstancesProcessor, _MarkdownTextAttributesProcessor):
                     self.paragraphs, self._md_text, "markdown"
                 )
 
-        return self
+    def _revalidate_after_deserialization(self) -> None:
+        """
+        Re-runs document consistency validation after private attributes have
+        been restored during deserialization.
+
+        ``from_dict`` constructs the instance from public fields (running model
+        validators) and only then restores private attributes such as
+        ``_md_text``. Because the document's markdown-consistency checks are
+        gated on ``_md_text`` being populated, they are skipped at construction
+        time during deserialization; this hook re-runs them (and re-normalizes
+        aspect nesting levels) once the full state is in place.
+
+        :return: None
+        :rtype: None
+        """
+        self._run_post_validation()

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/base/llms.py

@@ -136,7 +136,7 @@ if TYPE_CHECKING:
     )
 
 
-litellm.suppress_debug_info = True  # ty: ignore[invalid-assignment]
+litellm.suppress_debug_info = True
 
 
 # Local model providers supported via liteLLM
@@ -512,6 +512,11 @@ class _GenericLLMProcessor(_AbstractGenericLLMProcessor):
                 _Document,
                 paragraphs=aspect.reference_paragraphs,
             )
+            # Keep the sub-aspects' nesting levels intact: the extraction engine
+            # checks "aspect is assigned to document" via instance equality (which
+            # includes `_nesting_level`), so the added aspects must stay equal to
+            # the instances being processed rather than being normalized to 0.
+            aspect_document._suppress_aspect_nesting_normalization = True
             aspect_document.add_aspects(aspect.aspects)  # ty: ignore[invalid-argument-type]
             sub_aspect_jobs.append((aspect, aspect_document))
 
@@ -734,6 +739,11 @@ class _GenericLLMProcessor(_AbstractGenericLLMProcessor):
                     _Document,
                     paragraphs=sub_aspect.reference_paragraphs,
                 )
+                # Keep the sub-aspect's nesting level intact: the extraction engine
+                # checks "aspect is assigned to document" via instance equality
+                # (which includes `_nesting_level`), so the added aspect must stay
+                # equal to the instance being processed rather than normalized to 0.
+                sub_aspect_document._suppress_aspect_nesting_normalization = True
                 sub_aspect_document.add_aspects([sub_aspect])
                 sub_aspect_jobs.append((sub_aspect, sub_aspect_document))
 
@@ -3147,9 +3157,13 @@ class _DocumentLLM(_GenericLLMProcessor):
         description="Sampling temperature [0..1]; higher values increase randomness.",
     )
     top_p: StrictFloat | None = Field(
-        default=0.3,
+        default=None,
         ge=0,
-        description="Nucleus sampling [0..1]; alternative to temperature.",
+        description=(
+            "Nucleus sampling [0..1]; alternative to temperature. "
+            "Defaults to None so only temperature is sent — some providers "
+            "(e.g. Claude 4.x) reject requests that set both."
+        ),
     )
     seed: StrictInt | None = Field(
         default=None,
@@ -4064,23 +4078,9 @@ class _DocumentLLM(_GenericLLMProcessor):
                     "Fallback LLM must not have the exact same config params as the main LLM."
                 )
 
-        # "minimal" reasoning effort is supported only for gpt-5 models
-        if self.reasoning_effort == "minimal" and not (
-            self.model.startswith("azure/gpt-5")
-            or self.model.startswith("openai/gpt-5")
-        ):
-            raise ValueError(
-                "`reasoning_effort='minimal'` is supported only for gpt-5 models."
-            )
-
-        # "xhigh" reasoning effort is supported only for gpt-5.2 models
-        if self.reasoning_effort == "xhigh" and not (
-            self.model.startswith("azure/gpt-5.2")
-            or self.model.startswith("openai/gpt-5.2")
-        ):
-            raise ValueError(
-                "`reasoning_effort='xhigh'` is supported only for gpt-5.2 models."
-            )
+        # Note: model-specific support for `reasoning_effort` values is not
+        # validated here; validation is delegated to litellm / the provider API,
+        # which raises an error at request time for unsupported combinations.
 
         # Emit relevant warnings
 

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/base/serialization.py

@@ -92,6 +92,18 @@ KEY_RESPONSE_SUCCEEDED_PRIVATE = "_response_succeeded"
 KEY_MESSAGES_PRIVATE = "_messages"
 
 
+# Private attrs whose validation lives on a public property of the same name
+# (without the leading underscore). During deserialization these are assigned
+# through the public property setter so that its validation (e.g. element-type
+# and uniqueness checks) runs, instead of writing the backing attribute directly
+# and silently bypassing those invariants.
+_PRIVATE_ATTR_PUBLIC_PROPERTY: dict[str, str] = {
+    KEY_EXTRACTED_ITEMS_PRIVATE: "extracted_items",
+    KEY_REFERENCE_PARAGRAPHS_PRIVATE: "reference_paragraphs",
+    KEY_REFERENCE_SENTENCES_PRIVATE: "reference_sentences",
+}
+
+
 class _InstanceSerializer(BaseModel):
     """
     Base class that provides reusable methods for serialization and deserialization of instances.
@@ -516,12 +528,41 @@ class _InstanceSerializer(BaseModel):
 
         new_instance = cls(**constructor_kwargs)
 
-        # Set private attrs separately
+        # Set private attrs separately. Where a private attr is exposed through a
+        # public property carrying validation, assign via that property so the
+        # validation runs (instead of writing the backing attr directly and
+        # bypassing it); otherwise set the private attr directly.
         for priv_k, priv_v in private_attrs.items():
-            setattr(new_instance, priv_k, priv_v)
+            public_name = _PRIVATE_ATTR_PUBLIC_PROPERTY.get(priv_k)
+            if public_name is not None and isinstance(
+                getattr(type(new_instance), public_name, None), property
+            ):
+                setattr(new_instance, public_name, priv_v)
+            else:
+                setattr(new_instance, priv_k, priv_v)
+
+        # Re-run any validation that depends on private state restored above
+        # (and therefore could not run during construction from public fields).
+        new_instance._revalidate_after_deserialization()
 
         return new_instance
 
+    def _revalidate_after_deserialization(self) -> None:
+        """
+        Hook to re-run validation that depends on private attributes restored
+        after construction during :meth:`from_dict`.
+
+        Model validators run when the instance is constructed from its public
+        fields, before private attributes are restored, so any invariant gated
+        on private state is skipped at that point. Subclasses override this to
+        re-validate once the full state is in place. The base implementation is
+        a no-op.
+
+        :return: None
+        :rtype: None
+        """
+        return None
+
     @classmethod
     def _convert_llm_cost_dict(cls, cost_dict: dict[str, Any]) -> _LLMCost:
         """

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/prompts/extract_aspect_items.j2

@@ -195,7 +195,7 @@ ANSWER FORMAT (MANDATORY):
 [
     {
         "aspect_id": str,
-        "paragraphs: [
+        "paragraphs": [
             {
                 "paragraph_id": str,
     {% if reference_depth == "sentences" %}
@@ -220,7 +220,7 @@ ANSWER FORMAT (MANDATORY):
     {
         "aspect_id": str,
     {% if reference_depth == "sentences" %}
-        "paragraphs: [
+        "paragraphs": [
             {
                 "paragraph_id": str,
                 "sentences": [
@@ -233,7 +233,7 @@ ANSWER FORMAT (MANDATORY):
             ...
         ]
     {% else %}
-        "paragraph_ids: [
+        "paragraph_ids": [
             str,
             ...
         ]

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/tools/schema_generator.py

@@ -77,9 +77,37 @@ _JSON_SCHEMA_TYPE_MAP: dict[type, dict[str, Any]] = {
 }
 
 
+def _json_schema_type_of_value(value: Any) -> str | None:
+    """
+    Returns the JSON Schema ``type`` name for a concrete ``Literal`` value, or
+    ``None`` if the value has no single corresponding JSON Schema scalar type.
+
+    Booleans are checked before integers because ``bool`` is a subclass of
+    ``int``.
+
+    :param value: The literal value to classify.
+    :type value: Any
+    :returns: The JSON Schema type name (e.g. ``"integer"``), or ``None``.
+    :rtype: str | None
+    """
+    if isinstance(value, bool):
+        return "boolean"
+    if isinstance(value, int):
+        return "integer"
+    if isinstance(value, float):
+        return "number"
+    if isinstance(value, str):
+        return "string"
+    if value is None:
+        return "null"
+    return None
+
+
 def _python_type_to_json_schema(
     type_hint: Any,
     description: str | None = None,
+    *,
+    _seen: frozenset[int] = frozenset(),
 ) -> dict[str, Any]:
     """
     Convert a Python type hint to a JSON Schema object.
@@ -98,6 +126,11 @@ def _python_type_to_json_schema(
     :type type_hint: Any
     :param description: Optional description to include in the schema.
     :type description: str | None
+    :param _seen: Identities (``id()``) of class/TypedDict types currently being
+        expanded on the recursion stack. Used to break cycles in self- or
+        mutually-referential types; on revisit a generic ``{"type": "object"}``
+        is emitted instead of recursing without bound. Internal use only.
+    :type _seen: frozenset[int]
     :returns: JSON Schema object.
     :rtype: dict[str, Any]
     :raises TypeError: If the type hint is not supported.
@@ -126,7 +159,7 @@ def _python_type_to_json_schema(
         item_type = args[0] if args else Any
         schema = {
             "type": "array",
-            "items": _python_type_to_json_schema(item_type),
+            "items": _python_type_to_json_schema(item_type, _seen=_seen),
         }
         if description:
             schema["description"] = description
@@ -140,12 +173,14 @@ def _python_type_to_json_schema(
             if len(args) == 2 and args[1] is ...:
                 schema = {
                     "type": "array",
-                    "items": _python_type_to_json_schema(args[0]),
+                    "items": _python_type_to_json_schema(args[0], _seen=_seen),
                 }
             else:
                 schema = {
                     "type": "array",
-                    "prefixItems": [_python_type_to_json_schema(a) for a in args],
+                    "prefixItems": [
+                        _python_type_to_json_schema(a, _seen=_seen) for a in args
+                    ],
                     "minItems": len(args),
                     "maxItems": len(args),
                 }
@@ -160,7 +195,7 @@ def _python_type_to_json_schema(
         item_type = args[0] if args else Any
         schema = {
             "type": "array",
-            "items": _python_type_to_json_schema(item_type),
+            "items": _python_type_to_json_schema(item_type, _seen=_seen),
             "uniqueItems": True,
         }
         if description:
@@ -173,7 +208,9 @@ def _python_type_to_json_schema(
             value_type = args[1]
             schema = {
                 "type": "object",
-                "additionalProperties": _python_type_to_json_schema(value_type),
+                "additionalProperties": _python_type_to_json_schema(
+                    value_type, _seen=_seen
+                ),
             }
         else:
             schema = {"type": "object"}
@@ -193,10 +230,12 @@ def _python_type_to_json_schema(
         if len(non_none_types) == 1 and has_none:
             # This is Optional[X] - return schema for X
             # The "required" handling happens at the parameter level
-            return _python_type_to_json_schema(non_none_types[0], description)
+            return _python_type_to_json_schema(
+                non_none_types[0], description, _seen=_seen
+            )
 
         # Multiple types: use anyOf
-        schema = {"anyOf": [_python_type_to_json_schema(a) for a in args]}
+        schema = {"anyOf": [_python_type_to_json_schema(a, _seen=_seen) for a in args]}
         if description:
             schema["description"] = description
         return schema
@@ -207,14 +246,13 @@ def _python_type_to_json_schema(
         if not literal_values:
             raise TypeError("Literal type must have at least one value")
 
-        # Determine type from first value
-        first_val = literal_values[0]
-        if isinstance(first_val, bool):
-            schema = {"type": "boolean", "enum": literal_values}
-        elif isinstance(first_val, int):
-            schema = {"type": "integer", "enum": literal_values}
-        elif isinstance(first_val, str):
-            schema = {"type": "string", "enum": literal_values}
+        # Only assign a JSON Schema "type" when every literal value maps to the
+        # same scalar type; a mixed-type Literal (e.g. Literal[1, "a"]) would
+        # otherwise produce an invalid/mislabeled schema (a string under
+        # "type": "integer"), so emit a bare enum in that case.
+        value_types = {_json_schema_type_of_value(v) for v in literal_values}
+        if len(value_types) == 1 and None not in value_types:
+            schema = {"type": value_types.pop(), "enum": literal_values}
         else:
             schema = {"enum": literal_values}
 
@@ -224,15 +262,33 @@ def _python_type_to_json_schema(
 
     # Handle TypedDict
     if is_typeddict(type_hint):
+        # Break cycles in self-/mutually-referential TypedDicts
+        if id(type_hint) in _seen:
+            schema = {"type": "object"}
+            if description:
+                schema["description"] = description
+            return schema
         # type_hint is guaranteed to be a TypedDict type here
-        return _typeddict_to_json_schema(type_hint, description)  # type: ignore[arg-type]
+        return _typeddict_to_json_schema(
+            type_hint,  # type: ignore[arg-type]
+            description,
+            _seen | {id(type_hint)},
+        )
 
     # Handle plain classes that might be dict-like or have annotations
     if isinstance(type_hint, type):
         # If it's a basic class without special handling, treat as object
         if hasattr(type_hint, "__annotations__") and type_hint.__annotations__:
+            # Break cycles in self-/mutually-referential classes
+            if id(type_hint) in _seen:
+                schema = {"type": "object"}
+                if description:
+                    schema["description"] = description
+                return schema
             # Class with annotations - try to convert
-            return _class_to_json_schema(type_hint, description)
+            return _class_to_json_schema(
+                type_hint, description, _seen | {id(type_hint)}
+            )
 
         # Fallback: treat as generic object
         schema = {"type": "object"}
@@ -251,6 +307,7 @@ def _python_type_to_json_schema(
 def _typeddict_to_json_schema(
     td_type: type,
     description: str | None = None,
+    _seen: frozenset[int] = frozenset(),
 ) -> dict[str, Any]:
     """
     Convert a TypedDict to a JSON Schema object.
@@ -259,6 +316,9 @@ def _typeddict_to_json_schema(
     :type td_type: type
     :param description: Optional description to include in the schema.
     :type description: str | None
+    :param _seen: Identities of class/TypedDict types currently being expanded,
+        used for cycle detection. Internal use only.
+    :type _seen: frozenset[int]
     :returns: JSON Schema object.
     :rtype: dict[str, Any]
     """
@@ -269,7 +329,7 @@ def _typeddict_to_json_schema(
 
     properties: dict[str, Any] = {}
     for name, hint in hints.items():
-        properties[name] = _python_type_to_json_schema(hint)
+        properties[name] = _python_type_to_json_schema(hint, _seen=_seen)
 
     # Get required keys (TypedDict has __required_keys__ and __optional_keys__)
     required_keys = getattr(td_type, "__required_keys__", frozenset())
@@ -296,6 +356,7 @@ def _typeddict_to_json_schema(
 def _class_to_json_schema(
     cls_type: type,
     description: str | None = None,
+    _seen: frozenset[int] = frozenset(),
 ) -> dict[str, Any]:
     """
     Convert a class with annotations to a JSON Schema object.
@@ -304,6 +365,9 @@ def _class_to_json_schema(
     :type cls_type: type
     :param description: Optional description to include in the schema.
     :type description: str | None
+    :param _seen: Identities of class/TypedDict types currently being expanded,
+        used for cycle detection. Internal use only.
+    :type _seen: frozenset[int]
     :returns: JSON Schema object.
     :rtype: dict[str, Any]
     """
@@ -314,7 +378,7 @@ def _class_to_json_schema(
 
     properties: dict[str, Any] = {}
     for name, hint in hints.items():
-        properties[name] = _python_type_to_json_schema(hint)
+        properties[name] = _python_type_to_json_schema(hint, _seen=_seen)
 
     schema: dict[str, Any] = {
         "type": "object",

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/typings/strings_to_types.py

@@ -214,23 +214,34 @@ def _parse_type_hint(s: str, i: int = 0) -> tuple[Any, int]:
 
             # Parse the literal value (string or other)
             if i < len(s) and s[i] == '"':
-                # Parse string literal
-                i += 1  # skip opening quote
+                # Parse a double-quoted string literal, honoring backslash escapes.
+                # A backslash escapes the next character (so `\\` -> `\` and
+                # `\"` -> `"`); only an unescaped `"` terminates the literal. This
+                # is the inverse of `_serialize_literal_value`, which escapes `\`
+                # and `"` when emitting a string literal value.
                 start = i
-                # Find the closing quote, accounting for escaped quotes
+                i += 1  # skip opening quote
+                chars: list[str] = []
+                terminated = False
                 while i < len(s):
-                    if s[i] == '"' and (i == 0 or s[i - 1] != "\\"):
+                    ch = s[i]
+                    if ch == "\\" and i + 1 < len(s):
+                        # Consume the backslash and take the next character literally
+                        chars.append(s[i + 1])
+                        i += 2
+                        continue
+                    if ch == '"':
+                        terminated = True
                         break
+                    chars.append(ch)
                     i += 1
 
-                if i >= len(s):
+                if not terminated:
                     raise ValueError(
                         f"Unterminated string literal in literal type at position {start} in {s!r}"
                     )
 
-                # Get the string value and unescape quotes
-                string_value = s[start:i].replace('\\"', '"')
-                values.append(string_value)
+                values.append("".join(chars))
                 i += 1  # skip closing quote
 
             else:

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/typings/types.py

@@ -91,7 +91,7 @@ DefaultDecimalField = Field(
     default_factory=lambda: Decimal("0.00000"), ge=Decimal("0.00000")
 )
 
-ReasoningEffort = Literal["minimal", "low", "medium", "high", "xhigh"]
+ReasoningEffort = Literal["none", "minimal", "low", "medium", "high", "xhigh", "max"]
 
 TextMode = Literal["raw", "markdown"]
 

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/internal/typings/types_to_strings.py

@@ -49,6 +49,42 @@ PRIMITIVE_TYPES_STRING_MAP = {
 }
 
 
+def _serialize_literal_value(value: Any) -> str:
+    """
+    Serializes a single ``Literal`` value to its canonical string token.
+
+    The output is the inverse of the literal-value parsing performed during
+    deserialization (``strings_to_types._parse_type_hint``), so that
+    ``Literal`` round-trips losslessly:
+
+    - ``str`` -> a double-quoted string with ``\\`` and ``"`` escaped
+    - ``True`` / ``False`` -> ``"true"`` / ``"false"`` (lowercase JSON tokens)
+    - ``None`` -> ``"null"``
+    - ``int`` / ``float`` -> their plain string representation
+
+    Booleans are checked before the ``str`` branch is reached and before any
+    numeric handling, since ``bool`` is a subclass of ``int``.
+
+    :param value: The literal value to serialize.
+    :type value: Any
+    :return: The canonical string token for the literal value.
+    :rtype: str
+    """
+    if isinstance(value, str):
+        # Escape backslashes first, then double quotes, and wrap in double quotes.
+        # Order matters: escaping quotes first would double-escape the backslashes
+        # introduced by the quote escaping.
+        escaped = value.replace("\\", "\\\\").replace('"', '\\"')
+        return f'"{escaped}"'
+    if value is True:
+        return "true"
+    if value is False:
+        return "false"
+    if value is None:
+        return "null"
+    return str(value)
+
+
 def _raise_json_serializable_type_error(
     invalid_type: Any,
     field_name: str | None = None,
@@ -243,14 +279,7 @@ def _format_type(typ: Any, indent_level: int = 0) -> str:
             if getattr(arg, "__origin__", None) is Literal:
                 # Extract individual literal values
                 for literal_arg in get_args(arg):
-                    if isinstance(literal_arg, str):
-                        flattened_values.append(
-                            '"' + literal_arg.replace('"', '\\"') + '"'
-                        )
-                    elif literal_arg is None:
-                        flattened_values.append("null")
-                    else:
-                        flattened_values.append(str(literal_arg))
+                    flattened_values.append(_serialize_literal_value(literal_arg))
             elif arg is type(None):
                 flattened_values.append("null")
             else:
@@ -268,17 +297,7 @@ def _format_type(typ: Any, indent_level: int = 0) -> str:
     # Handle Literal types
     if getattr(typ, "__origin__", None) is Literal:
         # Create a list of serialized literal values
-        literal_values = []
-        for arg in get_args(typ):
-            if isinstance(arg, str):
-                # Escape quotes in strings and wrap in quotes
-                literal_values.append('"' + arg.replace('"', '\\"') + '"')
-            elif arg is None:
-                # Handle None consistently with standalone None type
-                literal_values.append("null")
-            else:
-                # For non-string literals, just convert to string
-                literal_values.append(str(arg))
+        literal_values = [_serialize_literal_value(arg) for arg in get_args(typ)]
 
         # Join all literal values with "or" instead of commas
         values_str = " or ".join(literal_values)
@@ -468,14 +487,7 @@ def _serialize_type_hint(tp: Any) -> str:
     # Handle Literal types
     elif origin is Literal:
         # Create a list of serialized literal values
-        literal_values = []
-        for arg in args:
-            if isinstance(arg, str):
-                # Escape quotes in strings and wrap in quotes
-                literal_values.append('"' + arg.replace('"', '\\"') + '"')
-            else:
-                # For non-string literals, just convert to string
-                literal_values.append(str(arg))
+        literal_values = [_serialize_literal_value(arg) for arg in args]
 
         # Join all literal values with commas
         values_str = ", ".join(literal_values)

{contextgem-0.25.0 → contextgem-0.26.0}/contextgem/public/llms.py

@@ -103,12 +103,17 @@ class DocumentLLM(_DocumentLLM):
         (CoT-capable) models. Defaults to 16000.
     :vartype max_completion_tokens: int
     :ivar reasoning_effort: The effort level for the LLM to reason about the input. Can be set to
-        ``"minimal"`` (gpt-5 models only), ``"low"``, ``"medium"``, ``"high"``, or ``"xhigh"``
-        (gpt-5.2 models only). Relevant for reasoning (CoT-capable) models. Defaults to None.
+        ``"none"``, ``"minimal"``, ``"low"``, ``"medium"``, ``"high"``, ``"xhigh"``, or
+        ``"max"``. Support for specific values varies by model and provider; unsupported
+        combinations are rejected by litellm or the provider API at request time. Note that
+        ``"none"`` explicitly disables reasoning on models that support this value, which
+        differs from the default ``None`` (parameter not sent, provider default applies).
+        Relevant for reasoning (CoT-capable) models. Defaults to None.
     :vartype reasoning_effort: ReasoningEffort | None
     :ivar top_p: Nucleus sampling value (0.0 to 1.0) controlling output focus/randomness.
         Lower values make output more deterministic, higher values produce more diverse outputs.
-        Defaults to 0.3.
+        Defaults to ``None`` (only ``temperature`` is sent). Some providers (e.g. Claude 4.x)
+        reject requests that set both ``temperature`` and ``top_p``.
     :vartype top_p: float | None
     :ivar num_retries_failed_request: Number of retries when LLM request fails. Defaults to 3.
     :vartype num_retries_failed_request: int

{contextgem-0.25.0 → contextgem-0.26.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "contextgem"
-version = "0.25.0"
+version = "0.26.0"
 description = "Effortless LLM extraction from documents"
 authors = [{ name = "shcherbak-ai", email = "sergii@shcherbak.ai" }]
 requires-python = ">=3.10,<3.15"
@@ -96,11 +96,11 @@ dependencies = [
     "colorlog (>=6.9.0,<7.0.0)",
     "docstring-parser (>=0.17.0,<1.0.0)",
     "fastjsonschema (>=2.21.2,<3.0.0)",
-    "genai-prices==0.0.59", # pin version due to frequent API changes
+    "genai-prices==0.0.72", # pin version due to frequent API changes
     "jinja2 (>=3.1.5,<4.0.0)",
-    "litellm==1.83.14", # pin version due to occasional instability between patches and minor versions
+    "litellm==1.93.0", # pin version due to occasional instability between patches and minor versions
     "lxml (>=5.4.0,<7.0.0)",
-    "openai==2.24.0", # pinned version for stable compatibility with litellm (after issue with 1.100.0)
+    "openai==2.49.0", # pinned version for stable compatibility with litellm (after issue with 1.100.0)
     "pillow (>=11.3.0,<12.0.0)",
     "pydantic (>=2.10.6,<3.0.0)",
     "python-ulid (>=3.0.0,<4.0.0)",
@@ -132,12 +132,12 @@ dev = [
     "nest-asyncio (>=1.6.0,<2.0.0)",
     "pre-commit (>=4.1.0,<5.0.0)",
     "pympler (>=1.1.0,<2.0.0)",
-    "ty (>=0.0.34,<1.0.0)",
+    "ty (==0.0.35)", # pinned to the last fully-clean ty release; newer ones surface issues here; unpin when issues are fixed
     "pytest (>=8.3.4,<9.0.0)",
     "pytest-cov (>=6.0.0,<7.0.0)",
     "pytest-recording (>=0.13.4,<1.0.0)",
     "python-dotenv (>=1.0.1,<2.0.0)",
-    "ruff (>=0.12.3,<1.0.0)",
+    "ruff (>=0.12.3,<0.16.0)", # ruff 0.16 expands the default lint rule set, flagging many new issues; capped until a dedicated lint migration
     "sphinx (>=7.0.0,<8.0.0)",
     "sphinx-autodoc-typehints<3.0.0",
     "sphinx-autobuild (>=2024.10.3,<2027.0.0)",
@@ -148,6 +148,7 @@ dev = [
     "sphinxext-opengraph (>=0.9.1,<1.0.0)",
     "types-lxml (>=2025.3.30,<2027.0.0)",
     "twine>=6.2.0",
+    "licenseal>=0.1.1",
 ]
 
 [build-system]
@@ -211,7 +212,7 @@ log_cli_level = "WARNING"
 
 [tool.commitizen]
 name = "cz_conventional_commits"
-version = "0.25.0"
+version = "0.26.0"
 version_files = [
     "contextgem/__init__.py:__version__",
     "pyproject.toml:version",