langchain-halu 0.1.0__tar.gz

langchain_halu-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Komplex AI
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

langchain_halu-0.1.0/PKG-INFO

@@ -0,0 +1,172 @@
+Metadata-Version: 2.4
+Name: langchain-halu
+Version: 0.1.0
+Summary: Thin LangChain integration for Komplex AI — a hallucination detector for LLM output (adapts the halu SDK to LCEL).
+Author-email: Komplex AI <support@komplexai.io>
+Maintainer-email: Komplex AI <support@komplexai.io>
+License: Apache-2.0
+Project-URL: Homepage, https://detector.komplexai.io
+Project-URL: Documentation, https://detector.komplexai.io/guide
+Project-URL: Source, https://github.com/komplexai
+Keywords: langchain,hallucination,hallucination-detection,llm,uncertainty,komplexai,halu,guardrails,runnable,callback
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Typing :: Typed
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: halu>=0.1.2
+Requires-Dist: langchain-core>=0.3
+Provides-Extra: test
+Requires-Dist: pytest>=7.0; extra == "test"
+Dynamic: license-file
+
+# langchain-halu
+
+A thin [LangChain](https://python.langchain.com/) integration for
+[**Komplex AI**](https://detector.komplexai.io) — a hallucination detector for
+LLM output. It adapts the [`halu`](https://pypi.org/project/halu/) Python SDK to
+LangChain so you can annotate, guard, monitor, or retry generations from inside
+an LCEL pipeline.
+
+## Install
+
+```bash
+pip install langchain-halu
+```
+
+This pulls in `halu` and `langchain-core` (not the full `langchain`).
+
+Set your API key in the environment:
+
+```bash
+export HALU_API_KEY="sk-..."
+```
+
+Get a key at <https://detector.komplexai.io>.
+
+## Three patterns
+
+### 1. Annotate (non-blocking)
+
+Enrich an answer with a verdict; the answer always passes through.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_openai import ChatOpenAI  # any LangChain chat model works
+from langchain_halu import halu_annotate
+
+llm = ChatOpenAI(model="gpt-4o-mini")
+chain = llm | StrOutputParser() | halu_annotate()
+
+result = chain.invoke("Who painted the Mona Lisa?")
+# {
+#   "answer": "Leonardo da Vinci painted the Mona Lisa.",
+#   "p_hallucination": 0.02,
+#   "flag": False,
+#   "top_regime": "NORMAL",
+# }
+```
+
+### 2. Guard (blocking)
+
+Reject flagged answers so they never reach the user.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import halu_guard, HaluHallucinationFlagged
+
+chain = llm | StrOutputParser() | halu_guard(threshold=0.5)
+
+try:
+    answer = chain.invoke("Summarize the plot of a book that does not exist.")
+except HaluHallucinationFlagged as exc:
+    result = exc.detection_result
+    print(f"Blocked (p={result.p_hallucination:.2f}, regime={result.top_regime})")
+    answer = "Sorry, I couldn't produce a reliable answer."
+```
+
+### 3. Monitor (passive callback)
+
+Observe every completion and log flagged answers, without changing behavior.
+
+```python
+from langchain_halu import HaluCallbackHandler
+
+monitor = HaluCallbackHandler(threshold=0.5)
+llm.invoke("What year did the Great Fire of London happen?", config={"callbacks": [monitor]})
+
+if monitor.last_result and monitor.last_result.flag:
+    print("Flagged:", monitor.last_result.top_regime)
+# monitor.flagged_results holds every flagged DetectResult seen so far.
+```
+
+### Bonus: Retry until clean
+
+Regenerate with regime feedback until the answer passes the detector.
+
+```python
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import generate_clean
+
+prompt = ChatPromptTemplate.from_template("{input}")
+chain = prompt | llm | StrOutputParser()
+
+answer = generate_clean(
+    chain,
+    prompt="Who discovered penicillin, and in what year?",
+    max_retries=2,
+    threshold=0.5,
+)
+```
+
+`generate_clean` accepts any LangChain `Runnable` (anything with `.invoke`) or a
+plain callable that takes a prompt string and returns text or an
+`AIMessage`-like object.
+
+## API
+
+| Name | Kind | Purpose |
+|------|------|---------|
+| `halu_annotate(prompt=None)` | `Runnable` factory | Return `{"answer", "p_hallucination", "flag", "top_regime"}` |
+| `halu_guard(threshold=0.5, prompt=None)` | `Runnable` factory | Pass the answer through or raise `HaluHallucinationFlagged` |
+| `HaluCallbackHandler(threshold=0.5, prompt=None, log_level=WARNING)` | Callback | Detect on `on_llm_end`, log + store flagged results |
+| `generate_clean(chain_or_callable, prompt, max_retries=2, threshold=0.5)` | Function | Retry generation until clean |
+
+`DetectResult`, `HaluError`, and `HaluHallucinationFlagged` are re-exported from
+`halu` for convenience.
+
+## Configuration
+
+Detection reads `HALU_API_KEY` from the environment by default and talks to
+`https://api.komplexai.io`. All the standard `halu` options (`api_key`,
+`base_url`, `timeout`) are honored through `halu`'s own configuration; see the
+[`halu` docs](https://pypi.org/project/halu/).
+
+## Scope & limits
+
+The detector is optimized for a specific slice, and results outside it are not
+reliable:
+
+- **English, natural-language** responses only.
+- Responses up to **2048 characters**.
+- The **first call after an idle period can cold-start for ~10–30 s**; subsequent
+  calls are fast. Build a warm-up call into latency-sensitive apps.
+- Regimes reported: `NORMAL`, `FABRICATED`, `NEAR_FALSE`, `CF_AUTH` (fake or
+  misattributed citation), `FALSE_REFUSAL`, `Other`.
+
+Full guidance and best practices: <https://detector.komplexai.io/guide>.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

langchain_halu-0.1.0/README.md

@@ -0,0 +1,140 @@
+# langchain-halu
+
+A thin [LangChain](https://python.langchain.com/) integration for
+[**Komplex AI**](https://detector.komplexai.io) — a hallucination detector for
+LLM output. It adapts the [`halu`](https://pypi.org/project/halu/) Python SDK to
+LangChain so you can annotate, guard, monitor, or retry generations from inside
+an LCEL pipeline.
+
+## Install
+
+```bash
+pip install langchain-halu
+```
+
+This pulls in `halu` and `langchain-core` (not the full `langchain`).
+
+Set your API key in the environment:
+
+```bash
+export HALU_API_KEY="sk-..."
+```
+
+Get a key at <https://detector.komplexai.io>.
+
+## Three patterns
+
+### 1. Annotate (non-blocking)
+
+Enrich an answer with a verdict; the answer always passes through.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_openai import ChatOpenAI  # any LangChain chat model works
+from langchain_halu import halu_annotate
+
+llm = ChatOpenAI(model="gpt-4o-mini")
+chain = llm | StrOutputParser() | halu_annotate()
+
+result = chain.invoke("Who painted the Mona Lisa?")
+# {
+#   "answer": "Leonardo da Vinci painted the Mona Lisa.",
+#   "p_hallucination": 0.02,
+#   "flag": False,
+#   "top_regime": "NORMAL",
+# }
+```
+
+### 2. Guard (blocking)
+
+Reject flagged answers so they never reach the user.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import halu_guard, HaluHallucinationFlagged
+
+chain = llm | StrOutputParser() | halu_guard(threshold=0.5)
+
+try:
+    answer = chain.invoke("Summarize the plot of a book that does not exist.")
+except HaluHallucinationFlagged as exc:
+    result = exc.detection_result
+    print(f"Blocked (p={result.p_hallucination:.2f}, regime={result.top_regime})")
+    answer = "Sorry, I couldn't produce a reliable answer."
+```
+
+### 3. Monitor (passive callback)
+
+Observe every completion and log flagged answers, without changing behavior.
+
+```python
+from langchain_halu import HaluCallbackHandler
+
+monitor = HaluCallbackHandler(threshold=0.5)
+llm.invoke("What year did the Great Fire of London happen?", config={"callbacks": [monitor]})
+
+if monitor.last_result and monitor.last_result.flag:
+    print("Flagged:", monitor.last_result.top_regime)
+# monitor.flagged_results holds every flagged DetectResult seen so far.
+```
+
+### Bonus: Retry until clean
+
+Regenerate with regime feedback until the answer passes the detector.
+
+```python
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import generate_clean
+
+prompt = ChatPromptTemplate.from_template("{input}")
+chain = prompt | llm | StrOutputParser()
+
+answer = generate_clean(
+    chain,
+    prompt="Who discovered penicillin, and in what year?",
+    max_retries=2,
+    threshold=0.5,
+)
+```
+
+`generate_clean` accepts any LangChain `Runnable` (anything with `.invoke`) or a
+plain callable that takes a prompt string and returns text or an
+`AIMessage`-like object.
+
+## API
+
+| Name | Kind | Purpose |
+|------|------|---------|
+| `halu_annotate(prompt=None)` | `Runnable` factory | Return `{"answer", "p_hallucination", "flag", "top_regime"}` |
+| `halu_guard(threshold=0.5, prompt=None)` | `Runnable` factory | Pass the answer through or raise `HaluHallucinationFlagged` |
+| `HaluCallbackHandler(threshold=0.5, prompt=None, log_level=WARNING)` | Callback | Detect on `on_llm_end`, log + store flagged results |
+| `generate_clean(chain_or_callable, prompt, max_retries=2, threshold=0.5)` | Function | Retry generation until clean |
+
+`DetectResult`, `HaluError`, and `HaluHallucinationFlagged` are re-exported from
+`halu` for convenience.
+
+## Configuration
+
+Detection reads `HALU_API_KEY` from the environment by default and talks to
+`https://api.komplexai.io`. All the standard `halu` options (`api_key`,
+`base_url`, `timeout`) are honored through `halu`'s own configuration; see the
+[`halu` docs](https://pypi.org/project/halu/).
+
+## Scope & limits
+
+The detector is optimized for a specific slice, and results outside it are not
+reliable:
+
+- **English, natural-language** responses only.
+- Responses up to **2048 characters**.
+- The **first call after an idle period can cold-start for ~10–30 s**; subsequent
+  calls are fast. Build a warm-up call into latency-sensitive apps.
+- Regimes reported: `NORMAL`, `FABRICATED`, `NEAR_FALSE`, `CF_AUTH` (fake or
+  misattributed citation), `FALSE_REFUSAL`, `Other`.
+
+Full guidance and best practices: <https://detector.komplexai.io/guide>.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

langchain_halu-0.1.0/pyproject.toml

@@ -0,0 +1,63 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "langchain-halu"
+version = "0.1.0"
+description = "Thin LangChain integration for Komplex AI — a hallucination detector for LLM output (adapts the halu SDK to LCEL)."
+readme = "README.md"
+requires-python = ">=3.9"
+license = { text = "Apache-2.0" }
+authors = [
+    { name = "Komplex AI", email = "support@komplexai.io" },
+]
+maintainers = [
+    { name = "Komplex AI", email = "support@komplexai.io" },
+]
+keywords = [
+    "langchain",
+    "hallucination",
+    "hallucination-detection",
+    "llm",
+    "uncertainty",
+    "komplexai",
+    "halu",
+    "guardrails",
+    "runnable",
+    "callback",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Typing :: Typed",
+]
+dependencies = [
+    "halu>=0.1.2",
+    "langchain-core>=0.3",
+]
+
+[project.urls]
+Homepage = "https://detector.komplexai.io"
+Documentation = "https://detector.komplexai.io/guide"
+Source = "https://github.com/komplexai"
+
+[project.optional-dependencies]
+test = [
+    "pytest>=7.0",
+]
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+langchain_halu = ["py.typed"]

langchain_halu-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

langchain_halu-0.1.0/src/langchain_halu/__init__.py

@@ -0,0 +1,46 @@
+"""langchain-halu: a thin LangChain integration for Komplex AI — a hallucination detector for LLM output.
+
+This package adapts the ``halu`` SDK (Komplex AI's hallucination detector) to
+LangChain, exposing three complementary patterns:
+
+* :func:`halu_annotate` -- a Runnable that enriches an answer with a verdict
+  (non-blocking "annotate").
+* :func:`halu_guard` -- a Runnable that passes clean answers through and raises
+  on flagged ones (blocking "guard").
+* :class:`HaluCallbackHandler` -- a callback that passively monitors LLM
+  completions and logs flagged answers (observability "monitor").
+* :func:`generate_clean` -- regenerate an answer until it passes the detector
+  (the "retry" pattern).
+
+The ``halu`` error and result types are re-exported for convenience so callers
+do not need a separate ``import halu`` just to catch a flagged answer.
+
+Set ``HALU_API_KEY`` in the environment (or pass ``api_key`` through ``halu``)
+before use. See https://detector.komplexai.io/guide for details and limits.
+"""
+
+from __future__ import annotations
+
+from halu import (
+    DetectResult,
+    HaluError,
+    HaluHallucinationFlagged,
+)
+
+from .callbacks import HaluCallbackHandler
+from .retry import generate_clean
+from .runnables import halu_annotate, halu_guard
+
+__version__ = "0.1.0"
+
+__all__ = [
+    "halu_annotate",
+    "halu_guard",
+    "HaluCallbackHandler",
+    "generate_clean",
+    # Re-exported from halu for convenience.
+    "DetectResult",
+    "HaluError",
+    "HaluHallucinationFlagged",
+    "__version__",
+]

langchain_halu-0.1.0/src/langchain_halu/_common.py

@@ -0,0 +1,85 @@
+"""Shared helpers for the langchain-halu integration.
+
+This module isolates the small amount of glue logic used across the
+Runnable factories, the callback handler, and the retry helper: importing
+``langchain_core`` with a friendly error, and normalizing the many shapes a
+LangChain LLM step can hand us into a plain string.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+def require_langchain_core() -> None:
+    """Raise a clear, actionable error if ``langchain-core`` is not installed.
+
+    ``langchain-core`` is a declared dependency, but users occasionally end up
+    with a broken environment (e.g. after a partial install). Surfacing a
+    targeted message beats a bare ``ModuleNotFoundError`` from deep in an
+    import chain.
+    """
+    try:
+        import langchain_core  # noqa: F401
+    except ImportError as exc:  # pragma: no cover - environment-dependent
+        raise ImportError(
+            "langchain-halu requires 'langchain-core' (>=0.3). "
+            "Install it with `pip install langchain-core` "
+            "or `pip install langchain-halu`."
+        ) from exc
+
+
+def to_text(output: Any) -> str:
+    """Normalize a LangChain LLM/chain output into a plain string.
+
+    Handles the common shapes seen at the tail of a LangChain pipeline:
+
+    * a plain ``str`` (e.g. from ``StrOutputParser``);
+    * a message-like object exposing ``.content`` (e.g. ``AIMessage``),
+      whose content may itself be a string or a list of content blocks;
+    * anything else, which is coerced with ``str()`` as a last resort.
+
+    Parameters
+    ----------
+    output:
+        The value produced by an upstream LLM or chain step.
+
+    Returns
+    -------
+    str
+        The best-effort text representation of ``output``.
+    """
+    if isinstance(output, str):
+        return output
+
+    content = getattr(output, "content", None)
+    if content is not None:
+        return _content_to_text(content)
+
+    return str(output)
+
+
+def _content_to_text(content: Any) -> str:
+    """Flatten a message ``.content`` payload into a string.
+
+    Modern LangChain messages may carry ``content`` as a list of content
+    blocks (dicts with a ``"text"`` key, or already-plain strings) instead of
+    a single string. We concatenate the textual parts and ignore non-text
+    blocks (e.g. images/tool calls), which are not meaningful to a
+    natural-language hallucination detector.
+    """
+    if isinstance(content, str):
+        return content
+
+    if isinstance(content, (list, tuple)):
+        parts = []
+        for block in content:
+            if isinstance(block, str):
+                parts.append(block)
+            elif isinstance(block, dict):
+                text = block.get("text")
+                if isinstance(text, str):
+                    parts.append(text)
+        return "".join(parts)
+
+    return str(content)

langchain_halu-0.1.0/src/langchain_halu/callbacks.py

@@ -0,0 +1,117 @@
+"""LangChain callback handler for passive hallucination monitoring.
+
+``HaluCallbackHandler`` observes LLM completions as they finish and runs the
+Komplex AI detector on each one, logging a warning (and stashing the last
+result) whenever an answer is flagged. It never blocks or alters generation —
+this is the "monitor" pattern, meant for observability, dashboards, and alerts.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, List, Optional
+
+import halu
+
+from ._common import require_langchain_core
+
+logger = logging.getLogger("langchain_halu")
+
+require_langchain_core()
+from langchain_core.callbacks import BaseCallbackHandler  # noqa: E402
+
+
+class HaluCallbackHandler(BaseCallbackHandler):
+    """Run hallucination detection on every LLM completion, passively.
+
+    On ``on_llm_end`` the handler extracts each generated text, runs
+    ``halu.detect`` on it, and — when the result is flagged — emits a warning
+    log record and stores the result. Generation is never interrupted; use
+    :func:`langchain_halu.halu_guard` if you need to block flagged answers.
+
+    Parameters
+    ----------
+    threshold:
+        Probability at or above which a warning is logged. The detector's own
+        ``.flag`` is authoritative for the flagged/monitor decision; this
+        threshold is used as an additional gate so callers can tighten
+        sensitivity without changing the SDK default. Defaults to 0.5.
+    prompt:
+        Optional static prompt/context passed to the detector for every call.
+        Most callers leave this ``None`` and rely on response-only detection,
+        since the callback does not receive the original prompt text.
+    log_level:
+        Logging level used for flagged answers. Defaults to ``logging.WARNING``.
+
+    Attributes
+    ----------
+    last_result:
+        The most recent :class:`halu.DetectResult`, or ``None`` if no
+        completion has been processed yet.
+    flagged_results:
+        A list of all flagged :class:`halu.DetectResult` objects seen so far.
+    """
+
+    def __init__(
+        self,
+        threshold: float = 0.5,
+        prompt: Optional[str] = None,
+        log_level: int = logging.WARNING,
+    ) -> None:
+        super().__init__()
+        self.threshold = threshold
+        self.prompt = prompt
+        self.log_level = log_level
+        self.last_result: Optional["halu.DetectResult"] = None
+        self.flagged_results: List["halu.DetectResult"] = []
+
+    def on_llm_end(self, response: Any, **kwargs: Any) -> None:
+        """Detect on each generated text when an LLM run completes.
+
+        Parameters
+        ----------
+        response:
+            A LangChain ``LLMResult``-like object exposing ``.generations``
+            (a list of lists of generation objects, each with ``.text``).
+        """
+        for text in self._iter_texts(response):
+            if not text:
+                continue
+            try:
+                result = halu.detect(text, prompt=self.prompt)
+            except halu.HaluError as exc:
+                # Monitoring must never break the host application; log and move on.
+                logger.warning("langchain-halu: detection failed: %s", exc)
+                continue
+
+            self.last_result = result
+            if result.flag or result.p_hallucination >= self.threshold:
+                self.flagged_results.append(result)
+                logger.log(
+                    self.log_level,
+                    "langchain-halu: possible hallucination flagged "
+                    "(p=%.3f, top_regime=%s, request_id=%s)",
+                    result.p_hallucination,
+                    result.top_regime,
+                    result.request_id,
+                )
+
+    @staticmethod
+    def _iter_texts(response: Any) -> List[str]:
+        """Best-effort extraction of generated texts from an ``LLMResult``."""
+        texts: List[str] = []
+        generations = getattr(response, "generations", None)
+        if not generations:
+            return texts
+        for batch in generations:
+            for generation in batch:
+                text = getattr(generation, "text", None)
+                if isinstance(text, str) and text:
+                    texts.append(text)
+                    continue
+                # Chat generations may carry the text on a nested message.
+                message = getattr(generation, "message", None)
+                content = getattr(message, "content", None)
+                if isinstance(content, str) and content:
+                    texts.append(content)
+        return texts

langchain_halu-0.1.0/src/langchain_halu/retry.py

@@ -0,0 +1,81 @@
+"""Retry helper: regenerate an answer until the detector considers it clean.
+
+This wraps ``halu.regenerate_until_clean`` so it can drive a LangChain chain or
+any plain callable. It is the "retry" pattern: ask the model, check the answer,
+and if it is flagged, ask again with feedback about the failing regime, up to a
+bounded number of attempts.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Callable, Union
+
+import halu
+
+from ._common import to_text
+
+
+def generate_clean(
+    chain_or_callable: Union[Callable[[str], Any], Any],
+    prompt: str,
+    max_retries: int = 2,
+    threshold: float = 0.5,
+) -> Any:
+    """Generate an answer and retry until it passes the detector.
+
+    Parameters
+    ----------
+    chain_or_callable:
+        Either a LangChain ``Runnable`` (anything exposing ``.invoke``) or a
+        plain callable. It is called with a single string argument (the prompt,
+        possibly augmented with regeneration feedback) and is expected to
+        return an answer as a ``str`` or an ``AIMessage``-like object.
+    prompt:
+        The prompt/question to send to the model. Also passed to the detector
+        as context to improve detection quality.
+    max_retries:
+        Maximum number of regeneration attempts after the first try.
+        Defaults to 2.
+    threshold:
+        Acceptance threshold; an answer with ``p_hallucination`` below this is
+        accepted. Defaults to 0.5.
+
+    Returns
+    -------
+    Any
+        Whatever ``halu.regenerate_until_clean`` returns — typically the
+        accepted answer text (see the ``halu`` documentation for the exact
+        shape and behavior when no clean answer is found within the retry
+        budget).
+
+    Notes
+    -----
+    The underlying ``ask`` callback normalizes chain output to text via
+    :func:`langchain_halu._common.to_text`, so both string-returning and
+    message-returning chains work.
+
+    Examples
+    --------
+    >>> chain = prompt_template | llm  # doctest: +SKIP
+    >>> answer = generate_clean(chain, "Who discovered penicillin?")  # doctest: +SKIP
+    """
+    if hasattr(chain_or_callable, "invoke"):
+        invoke = chain_or_callable.invoke
+    elif callable(chain_or_callable):
+        invoke = chain_or_callable
+    else:  # pragma: no cover - defensive
+        raise TypeError(
+            "chain_or_callable must be a LangChain Runnable (with .invoke) "
+            "or a callable."
+        )
+
+    def ask(augmented_prompt: str) -> str:
+        return to_text(invoke(augmented_prompt))
+
+    return halu.regenerate_until_clean(
+        ask,
+        prompt=prompt,
+        max_retries=max_retries,
+        acceptance_threshold=threshold,
+        feedback_detail="regime",
+    )

langchain_halu-0.1.0/src/langchain_halu/runnables.py

@@ -0,0 +1,107 @@
+"""LangChain ``Runnable`` factories for the Komplex AI hallucination detector.
+
+These are thin adapters over ``halu.detect``. They accept whatever an upstream
+LLM/chain step emits (a ``str`` or an ``AIMessage``-like object) and plug into a
+LangChain Expression Language (LCEL) pipeline via the ``|`` operator.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Dict, Optional
+
+import halu
+
+from ._common import require_langchain_core, to_text
+
+
+def halu_annotate(prompt: Optional[str] = None) -> "Any":
+    """Build a ``Runnable`` that annotates an LLM answer with a detection verdict.
+
+    The returned runnable takes an LLM output (a ``str`` or an object with a
+    ``.content`` attribute, such as ``AIMessage``) and returns a dict::
+
+        {
+            "answer": <the original answer text>,
+            "p_hallucination": <float 0-1>,
+            "flag": <bool>,
+            "top_regime": <str>,
+        }
+
+    This is the non-blocking "annotate" pattern: the answer always passes
+    through, enriched with the detector's verdict so downstream code (or a
+    human) can decide what to do.
+
+    Parameters
+    ----------
+    prompt:
+        Optional prompt/context string passed to ``halu.detect`` to improve
+        detection. If ``None``, response-only detection is used.
+
+    Returns
+    -------
+    Runnable[Any, Dict[str, Any]]
+        A LangChain runnable suitable for use in an LCEL pipeline.
+
+    Examples
+    --------
+    >>> from langchain_core.output_parsers import StrOutputParser
+    >>> chain = llm | StrOutputParser() | halu_annotate()  # doctest: +SKIP
+    """
+    require_langchain_core()
+    from langchain_core.runnables import RunnableLambda
+
+    def _annotate(output: Any) -> Dict[str, Any]:
+        answer = to_text(output)
+        result = halu.detect(answer, prompt=prompt)
+        return {
+            "answer": answer,
+            "p_hallucination": result.p_hallucination,
+            "flag": result.flag,
+            "top_regime": result.top_regime,
+        }
+
+    return RunnableLambda(_annotate)
+
+
+def halu_guard(threshold: float = 0.5, prompt: Optional[str] = None) -> "Any":
+    """Build a ``Runnable`` that gates an LLM answer on the detector's verdict.
+
+    The returned runnable takes an LLM output (a ``str`` or an object with a
+    ``.content`` attribute) and either:
+
+    * returns the answer text unchanged when it is not flagged, or
+    * raises :class:`halu.HaluHallucinationFlagged` when the estimated
+      probability of hallucination meets or exceeds ``threshold``.
+
+    This is the blocking "guard" pattern: a flagged answer should not reach the
+    end user. Wrap downstream calls in ``try/except halu.HaluHallucinationFlagged``
+    to handle the rejection (e.g. show a fallback message or retry).
+
+    Parameters
+    ----------
+    threshold:
+        Probability at or above which the answer is rejected. Defaults to 0.5.
+    prompt:
+        Optional prompt/context string passed to the detector.
+
+    Returns
+    -------
+    Runnable[Any, str]
+        A LangChain runnable that returns the answer text or raises.
+
+    Examples
+    --------
+    >>> chain = llm | StrOutputParser() | halu_guard(threshold=0.5)  # doctest: +SKIP
+    """
+    require_langchain_core()
+    from langchain_core.runnables import RunnableLambda
+
+    def _guard(output: Any) -> str:
+        answer = to_text(output)
+        # detect_or_raise applies the threshold and raises
+        # HaluHallucinationFlagged (with .detection_result) when flagged;
+        # it returns the DetectResult otherwise.
+        halu.detect_or_raise(answer, threshold=threshold, prompt=prompt)
+        return answer
+
+    return RunnableLambda(_guard)

langchain_halu-0.1.0/src/langchain_halu.egg-info/PKG-INFO

@@ -0,0 +1,172 @@
+Metadata-Version: 2.4
+Name: langchain-halu
+Version: 0.1.0
+Summary: Thin LangChain integration for Komplex AI — a hallucination detector for LLM output (adapts the halu SDK to LCEL).
+Author-email: Komplex AI <support@komplexai.io>
+Maintainer-email: Komplex AI <support@komplexai.io>
+License: Apache-2.0
+Project-URL: Homepage, https://detector.komplexai.io
+Project-URL: Documentation, https://detector.komplexai.io/guide
+Project-URL: Source, https://github.com/komplexai
+Keywords: langchain,hallucination,hallucination-detection,llm,uncertainty,komplexai,halu,guardrails,runnable,callback
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Typing :: Typed
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: halu>=0.1.2
+Requires-Dist: langchain-core>=0.3
+Provides-Extra: test
+Requires-Dist: pytest>=7.0; extra == "test"
+Dynamic: license-file
+
+# langchain-halu
+
+A thin [LangChain](https://python.langchain.com/) integration for
+[**Komplex AI**](https://detector.komplexai.io) — a hallucination detector for
+LLM output. It adapts the [`halu`](https://pypi.org/project/halu/) Python SDK to
+LangChain so you can annotate, guard, monitor, or retry generations from inside
+an LCEL pipeline.
+
+## Install
+
+```bash
+pip install langchain-halu
+```
+
+This pulls in `halu` and `langchain-core` (not the full `langchain`).
+
+Set your API key in the environment:
+
+```bash
+export HALU_API_KEY="sk-..."
+```
+
+Get a key at <https://detector.komplexai.io>.
+
+## Three patterns
+
+### 1. Annotate (non-blocking)
+
+Enrich an answer with a verdict; the answer always passes through.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_openai import ChatOpenAI  # any LangChain chat model works
+from langchain_halu import halu_annotate
+
+llm = ChatOpenAI(model="gpt-4o-mini")
+chain = llm | StrOutputParser() | halu_annotate()
+
+result = chain.invoke("Who painted the Mona Lisa?")
+# {
+#   "answer": "Leonardo da Vinci painted the Mona Lisa.",
+#   "p_hallucination": 0.02,
+#   "flag": False,
+#   "top_regime": "NORMAL",
+# }
+```
+
+### 2. Guard (blocking)
+
+Reject flagged answers so they never reach the user.
+
+```python
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import halu_guard, HaluHallucinationFlagged
+
+chain = llm | StrOutputParser() | halu_guard(threshold=0.5)
+
+try:
+    answer = chain.invoke("Summarize the plot of a book that does not exist.")
+except HaluHallucinationFlagged as exc:
+    result = exc.detection_result
+    print(f"Blocked (p={result.p_hallucination:.2f}, regime={result.top_regime})")
+    answer = "Sorry, I couldn't produce a reliable answer."
+```
+
+### 3. Monitor (passive callback)
+
+Observe every completion and log flagged answers, without changing behavior.
+
+```python
+from langchain_halu import HaluCallbackHandler
+
+monitor = HaluCallbackHandler(threshold=0.5)
+llm.invoke("What year did the Great Fire of London happen?", config={"callbacks": [monitor]})
+
+if monitor.last_result and monitor.last_result.flag:
+    print("Flagged:", monitor.last_result.top_regime)
+# monitor.flagged_results holds every flagged DetectResult seen so far.
+```
+
+### Bonus: Retry until clean
+
+Regenerate with regime feedback until the answer passes the detector.
+
+```python
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_halu import generate_clean
+
+prompt = ChatPromptTemplate.from_template("{input}")
+chain = prompt | llm | StrOutputParser()
+
+answer = generate_clean(
+    chain,
+    prompt="Who discovered penicillin, and in what year?",
+    max_retries=2,
+    threshold=0.5,
+)
+```
+
+`generate_clean` accepts any LangChain `Runnable` (anything with `.invoke`) or a
+plain callable that takes a prompt string and returns text or an
+`AIMessage`-like object.
+
+## API
+
+| Name | Kind | Purpose |
+|------|------|---------|
+| `halu_annotate(prompt=None)` | `Runnable` factory | Return `{"answer", "p_hallucination", "flag", "top_regime"}` |
+| `halu_guard(threshold=0.5, prompt=None)` | `Runnable` factory | Pass the answer through or raise `HaluHallucinationFlagged` |
+| `HaluCallbackHandler(threshold=0.5, prompt=None, log_level=WARNING)` | Callback | Detect on `on_llm_end`, log + store flagged results |
+| `generate_clean(chain_or_callable, prompt, max_retries=2, threshold=0.5)` | Function | Retry generation until clean |
+
+`DetectResult`, `HaluError`, and `HaluHallucinationFlagged` are re-exported from
+`halu` for convenience.
+
+## Configuration
+
+Detection reads `HALU_API_KEY` from the environment by default and talks to
+`https://api.komplexai.io`. All the standard `halu` options (`api_key`,
+`base_url`, `timeout`) are honored through `halu`'s own configuration; see the
+[`halu` docs](https://pypi.org/project/halu/).
+
+## Scope & limits
+
+The detector is optimized for a specific slice, and results outside it are not
+reliable:
+
+- **English, natural-language** responses only.
+- Responses up to **2048 characters**.
+- The **first call after an idle period can cold-start for ~10–30 s**; subsequent
+  calls are fast. Build a warm-up call into latency-sensitive apps.
+- Regimes reported: `NORMAL`, `FABRICATED`, `NEAR_FALSE`, `CF_AUTH` (fake or
+  misattributed citation), `FALSE_REFUSAL`, `Other`.
+
+Full guidance and best practices: <https://detector.komplexai.io/guide>.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

langchain_halu-0.1.0/src/langchain_halu.egg-info/SOURCES.txt

@@ -0,0 +1,15 @@
+LICENSE
+README.md
+pyproject.toml
+src/langchain_halu/__init__.py
+src/langchain_halu/_common.py
+src/langchain_halu/callbacks.py
+src/langchain_halu/py.typed
+src/langchain_halu/retry.py
+src/langchain_halu/runnables.py
+src/langchain_halu.egg-info/PKG-INFO
+src/langchain_halu.egg-info/SOURCES.txt
+src/langchain_halu.egg-info/dependency_links.txt
+src/langchain_halu.egg-info/requires.txt
+src/langchain_halu.egg-info/top_level.txt
+tests/test_smoke.py

langchain_halu-0.1.0/src/langchain_halu.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

langchain_halu-0.1.0/src/langchain_halu.egg-info/requires.txt

@@ -0,0 +1,5 @@
+halu>=0.1.2
+langchain-core>=0.3
+
+[test]
+pytest>=7.0

langchain_halu-0.1.0/src/langchain_halu.egg-info/top_level.txt

@@ -0,0 +1 @@
+langchain_halu

langchain_halu-0.1.0/tests/test_smoke.py

@@ -0,0 +1,111 @@
+"""Light smoke tests for langchain-halu.
+
+These verify that the package imports and that its runnables, callback handler,
+and retry helper construct correctly without any network access. The single
+live end-to-end test is skipped by default and only runs when
+``HALU_LIVE_TEST=1`` is set in the environment (and a real ``HALU_API_KEY`` is
+available).
+
+Run with: ``pytest``
+"""
+
+import logging
+import os
+
+import pytest
+
+# The package depends on halu and langchain-core; if either is missing from the
+# test environment, skip the whole module rather than error out.
+pytest.importorskip("halu")
+pytest.importorskip("langchain_core")
+
+import langchain_halu  # noqa: E402
+from langchain_halu._common import to_text  # noqa: E402
+
+
+class _FakeMessage:
+    """Minimal AIMessage-like object exposing ``.content``."""
+
+    def __init__(self, content):
+        self.content = content
+
+
+def test_public_exports():
+    for name in (
+        "halu_annotate",
+        "halu_guard",
+        "HaluCallbackHandler",
+        "generate_clean",
+        "DetectResult",
+        "HaluError",
+        "HaluHallucinationFlagged",
+    ):
+        assert hasattr(langchain_halu, name), name
+    assert langchain_halu.__version__ == "0.1.0"
+
+
+def test_to_text_str():
+    assert to_text("hello") == "hello"
+
+
+def test_to_text_message_content_str():
+    assert to_text(_FakeMessage("hi there")) == "hi there"
+
+
+def test_to_text_message_content_blocks():
+    msg = _FakeMessage([{"type": "text", "text": "a"}, "b", {"image": "x"}])
+    assert to_text(msg) == "ab"
+
+
+def test_to_text_fallback():
+    assert to_text(123) == "123"
+
+
+def test_halu_annotate_constructs():
+    runnable = langchain_halu.halu_annotate()
+    assert hasattr(runnable, "invoke")
+
+
+def test_halu_guard_constructs():
+    runnable = langchain_halu.halu_guard(threshold=0.7)
+    assert hasattr(runnable, "invoke")
+
+
+def test_callback_handler_constructs():
+    handler = langchain_halu.HaluCallbackHandler(threshold=0.6)
+    assert handler.last_result is None
+    assert handler.flagged_results == []
+    assert handler.threshold == 0.6
+
+
+def test_callback_handler_is_base_callback_handler():
+    from langchain_core.callbacks import BaseCallbackHandler
+
+    assert isinstance(langchain_halu.HaluCallbackHandler(), BaseCallbackHandler)
+
+
+def test_callback_iter_texts_empty_response():
+    class _EmptyResult:
+        generations = []
+
+    handler = langchain_halu.HaluCallbackHandler()
+    # Should not raise and should not call the network on empty generations.
+    handler.on_llm_end(_EmptyResult())
+    assert handler.last_result is None
+
+
+def test_generate_clean_type_error_on_bad_input():
+    with pytest.raises(TypeError):
+        langchain_halu.generate_clean(object(), "prompt")
+
+
+@pytest.mark.skipif(
+    os.environ.get("HALU_LIVE_TEST") != "1",
+    reason="live network test; set HALU_LIVE_TEST=1 and HALU_API_KEY to run",
+)
+def test_annotate_live():
+    runnable = langchain_halu.halu_annotate()
+    out = runnable.invoke("The Eiffel Tower is located in Berlin, Germany.")
+    assert set(out) == {"answer", "p_hallucination", "flag", "top_regime"}
+    assert 0.0 <= out["p_hallucination"] <= 1.0
+    logging.getLogger(__name__).info("live annotate result: %s", out)