langchain-exaone 0.1.0__tar.gz

langchain_exaone-0.1.0/.gitignore

@@ -0,0 +1,18 @@
+.DS_Store
+.env
+.env.*
+!.env.example
+.venv/
+__pycache__/
+*.py[cod]
+*.egg-info/
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+.coverage.*
+node_modules/
+dist/
+build/
+coverage/
+/docs/

langchain_exaone-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 EXAONE Integrations contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

langchain_exaone-0.1.0/PKG-INFO

@@ -0,0 +1,133 @@
+Metadata-Version: 2.5
+Name: langchain-exaone
+Version: 0.1.0
+Summary: LangChain chat model integration for the EXAONE API
+Project-URL: Repository, https://github.com/ykoh42/exaone-integrations
+License-Expression: MIT
+License-File: LICENSE
+Requires-Python: >=3.10
+Requires-Dist: httpx<1,>=0.27
+Requires-Dist: langchain-core<2,>=1.7.5
+Requires-Dist: pydantic<3,>=2
+Description-Content-Type: text/markdown
+
+# `langchain-exaone`
+
+Community-maintained LangChain chat model integration for the EXAONE API.
+
+This package connects LangChain applications to the K-EXAONE 2.5 Chat Completions API. It implements LangChain's `BaseChatModel` interface and supports synchronous and asynchronous calls, streaming, tool calls, structured-output binding, usage metadata, and EXAONE reasoning history.
+
+This project is not affiliated with or endorsed by LG AI Research or LangChain.
+
+## Install
+
+This package is not published to PyPI yet. From a checkout of this repository, install it with:
+
+```bash
+pip install ./packages/langchain-exaone-py
+```
+
+After the first release, install it from PyPI with:
+
+```bash
+pip install -U langchain-exaone
+```
+
+Set `EXAONE_API_KEY` in the environment, then create the model:
+
+```python
+from langchain_exaone import ChatEXAONE
+
+model = ChatEXAONE()
+message = model.invoke("안녕하세요. K-EXAONE을 소개해 주세요.")
+print(message.text)
+```
+
+The package does not load `.env` files. Use your application's secret-management setup to place the API key in the process environment, or pass `api_key` directly.
+
+## Serialization
+
+`ChatEXAONE` does not opt into LangChain object serialization. LangChain's deserializer only allows registered integration classes by default, and this standalone package is not registered in `langchain-core`. The provider does not modify LangChain's global serialization registry. Create the model from application configuration when starting your process; its API key is held in a `SecretStr` and omitted from the model representation.
+
+## Reasoning history
+
+EXAONE returns reasoning separately from final content. The provider exposes it on `AIMessage.additional_kwargs["reasoning_content"]` and includes it when an assistant message is sent back in later turns. Keep the original `AIMessage` in conversation history so reasoning and tool calls remain attached to their assistant turn.
+
+```python
+from langchain_core.messages import HumanMessage
+from langchain_exaone import ChatEXAONE
+
+model = ChatEXAONE(reasoning_effort="low")
+history = [HumanMessage("2 + 2는 얼마야?")]
+answer = model.invoke(history)
+history.append(answer)
+history.append(HumanMessage("방금 답을 확인해줘."))
+follow_up = model.invoke(history)
+```
+
+Use `model.astream(...)` for asynchronous streaming. Streaming chunks retain reasoning in `additional_kwargs`, while text remains in the normal message content field.
+
+LangChain's streaming interface represents one completion per run. If the API returns multiple choices, such as when `n` is greater than `1` in `model_kwargs`, `stream()` and `astream()` raise `ValueError` rather than merging separate candidates. Use `generate()` when you need multiple candidates.
+
+## Tools and structured output
+
+```python
+from langchain_core.tools import tool
+from langchain_exaone import ChatEXAONE
+
+
+@tool
+def get_weather(city: str) -> str:
+    """Get the current weather for a city."""
+    return f"Sunny in {city}."
+
+
+model = ChatEXAONE().bind_tools([get_weather], tool_choice="auto")
+response = model.invoke("서울 날씨를 알려줘.")
+```
+
+`with_structured_output()` uses LangChain's standard tool-based strategy. You can pass additional Chat Completions fields with `model_kwargs` or per-call keyword arguments. Their acceptance depends on the EXAONE endpoint.
+
+## Configuration
+
+| Option | Default | Description |
+| --- | --- | --- |
+| `model` | `k-exaone_v2.5` | EXAONE model ID |
+| `api_key` | `EXAONE_API_KEY` | API credential |
+| `base_url` | `https://api.exaone.ai/v1` | API base URL. Remote endpoints must use HTTPS; HTTP is allowed for loopback test servers. |
+| `temperature` | `1.0` | Recommended EXAONE sampling value |
+| `top_p` | `0.95` | Recommended EXAONE sampling value |
+| `top_k` | `-1` | Disable top-k filtering, as in the API default |
+| `presence_penalty` | `0.0` | Recommended EXAONE sampling value |
+| `max_tokens` | API default | Maximum generated tokens, including reasoning |
+| `reasoning_effort` | API default | `low`, `medium`, `high`, or `max` |
+| `preserve_thinking` | `true` | Keep earlier assistant reasoning in multi-turn agent conversations |
+| `chat_template_kwargs` | `{}` | Additional chat-template options; `preserve_thinking` is set from its dedicated option |
+| `max_retries` | `2` | Retries for transient API errors before output is returned; set to `0` to disable |
+| `timeout` | `None` | No provider-imposed HTTP timeout; pass a positive number to set one |
+
+The API always enables reasoning. The provider preserves the returned `reasoning_content` on assistant messages and sends it with later turns; `preserve_thinking=True` also enables EXAONE's recommended `chat_template_kwargs.preserve_thinking` behavior. Keep returned assistant messages intact when continuing a conversation.
+
+`max_tokens` is capped at the documented 524,288-token context limit. The provider cannot count an arbitrary prompt locally, so the API still validates the combined input and output token budget.
+
+The provider retries HTTP 408, 429, 500, 502, 503, and 504 errors up to two times by default; set `max_retries=0` to disable retries. It honors the longer valid `Retry-After` or `Retry-After-Ms` delay. If neither header is present, the first retry waits at least one second to respect the API's per-second rate limit; a `max_retry_delay` below one second disables that retry. It does not retry after streaming output has started.
+
+The EXAONE API is text-only. Image, audio, and other non-text input blocks are rejected explicitly.
+
+See the [K-EXAONE 2.5 API reference](https://docs.exaone.ai/ko/docs/k-exaone-2.5) for the model's request parameters and limits.
+
+## Examples
+
+See [`examples/basic.py`](examples/basic.py) for a runnable example. The package does not make API requests during import or construction.
+
+## Development
+
+```bash
+uv sync --group test --group lint
+uv run pytest
+uv run ruff check .
+uv run ruff format --check .
+uv build
+```
+
+Unit tests use synthetic responses and do not require API credentials.

langchain_exaone-0.1.0/README.md

@@ -0,0 +1,120 @@
+# `langchain-exaone`
+
+Community-maintained LangChain chat model integration for the EXAONE API.
+
+This package connects LangChain applications to the K-EXAONE 2.5 Chat Completions API. It implements LangChain's `BaseChatModel` interface and supports synchronous and asynchronous calls, streaming, tool calls, structured-output binding, usage metadata, and EXAONE reasoning history.
+
+This project is not affiliated with or endorsed by LG AI Research or LangChain.
+
+## Install
+
+This package is not published to PyPI yet. From a checkout of this repository, install it with:
+
+```bash
+pip install ./packages/langchain-exaone-py
+```
+
+After the first release, install it from PyPI with:
+
+```bash
+pip install -U langchain-exaone
+```
+
+Set `EXAONE_API_KEY` in the environment, then create the model:
+
+```python
+from langchain_exaone import ChatEXAONE
+
+model = ChatEXAONE()
+message = model.invoke("안녕하세요. K-EXAONE을 소개해 주세요.")
+print(message.text)
+```
+
+The package does not load `.env` files. Use your application's secret-management setup to place the API key in the process environment, or pass `api_key` directly.
+
+## Serialization
+
+`ChatEXAONE` does not opt into LangChain object serialization. LangChain's deserializer only allows registered integration classes by default, and this standalone package is not registered in `langchain-core`. The provider does not modify LangChain's global serialization registry. Create the model from application configuration when starting your process; its API key is held in a `SecretStr` and omitted from the model representation.
+
+## Reasoning history
+
+EXAONE returns reasoning separately from final content. The provider exposes it on `AIMessage.additional_kwargs["reasoning_content"]` and includes it when an assistant message is sent back in later turns. Keep the original `AIMessage` in conversation history so reasoning and tool calls remain attached to their assistant turn.
+
+```python
+from langchain_core.messages import HumanMessage
+from langchain_exaone import ChatEXAONE
+
+model = ChatEXAONE(reasoning_effort="low")
+history = [HumanMessage("2 + 2는 얼마야?")]
+answer = model.invoke(history)
+history.append(answer)
+history.append(HumanMessage("방금 답을 확인해줘."))
+follow_up = model.invoke(history)
+```
+
+Use `model.astream(...)` for asynchronous streaming. Streaming chunks retain reasoning in `additional_kwargs`, while text remains in the normal message content field.
+
+LangChain's streaming interface represents one completion per run. If the API returns multiple choices, such as when `n` is greater than `1` in `model_kwargs`, `stream()` and `astream()` raise `ValueError` rather than merging separate candidates. Use `generate()` when you need multiple candidates.
+
+## Tools and structured output
+
+```python
+from langchain_core.tools import tool
+from langchain_exaone import ChatEXAONE
+
+
+@tool
+def get_weather(city: str) -> str:
+    """Get the current weather for a city."""
+    return f"Sunny in {city}."
+
+
+model = ChatEXAONE().bind_tools([get_weather], tool_choice="auto")
+response = model.invoke("서울 날씨를 알려줘.")
+```
+
+`with_structured_output()` uses LangChain's standard tool-based strategy. You can pass additional Chat Completions fields with `model_kwargs` or per-call keyword arguments. Their acceptance depends on the EXAONE endpoint.
+
+## Configuration
+
+| Option | Default | Description |
+| --- | --- | --- |
+| `model` | `k-exaone_v2.5` | EXAONE model ID |
+| `api_key` | `EXAONE_API_KEY` | API credential |
+| `base_url` | `https://api.exaone.ai/v1` | API base URL. Remote endpoints must use HTTPS; HTTP is allowed for loopback test servers. |
+| `temperature` | `1.0` | Recommended EXAONE sampling value |
+| `top_p` | `0.95` | Recommended EXAONE sampling value |
+| `top_k` | `-1` | Disable top-k filtering, as in the API default |
+| `presence_penalty` | `0.0` | Recommended EXAONE sampling value |
+| `max_tokens` | API default | Maximum generated tokens, including reasoning |
+| `reasoning_effort` | API default | `low`, `medium`, `high`, or `max` |
+| `preserve_thinking` | `true` | Keep earlier assistant reasoning in multi-turn agent conversations |
+| `chat_template_kwargs` | `{}` | Additional chat-template options; `preserve_thinking` is set from its dedicated option |
+| `max_retries` | `2` | Retries for transient API errors before output is returned; set to `0` to disable |
+| `timeout` | `None` | No provider-imposed HTTP timeout; pass a positive number to set one |
+
+The API always enables reasoning. The provider preserves the returned `reasoning_content` on assistant messages and sends it with later turns; `preserve_thinking=True` also enables EXAONE's recommended `chat_template_kwargs.preserve_thinking` behavior. Keep returned assistant messages intact when continuing a conversation.
+
+`max_tokens` is capped at the documented 524,288-token context limit. The provider cannot count an arbitrary prompt locally, so the API still validates the combined input and output token budget.
+
+The provider retries HTTP 408, 429, 500, 502, 503, and 504 errors up to two times by default; set `max_retries=0` to disable retries. It honors the longer valid `Retry-After` or `Retry-After-Ms` delay. If neither header is present, the first retry waits at least one second to respect the API's per-second rate limit; a `max_retry_delay` below one second disables that retry. It does not retry after streaming output has started.
+
+The EXAONE API is text-only. Image, audio, and other non-text input blocks are rejected explicitly.
+
+See the [K-EXAONE 2.5 API reference](https://docs.exaone.ai/ko/docs/k-exaone-2.5) for the model's request parameters and limits.
+
+## Examples
+
+See [`examples/basic.py`](examples/basic.py) for a runnable example. The package does not make API requests during import or construction.
+
+## Development
+
+```bash
+uv sync --group test --group lint
+uv run pytest
+uv run ruff check .
+uv run ruff format --check .
+uv build
+```
+
+Unit tests use synthetic responses and do not require API credentials.

langchain_exaone-0.1.0/examples/basic.py

@@ -0,0 +1,10 @@
+"""Call EXAONE with the LangChain chat model integration."""
+
+from langchain_exaone import ChatEXAONE
+
+model = ChatEXAONE(reasoning_effort="low")
+response = model.invoke("K-EXAONE을 한 문장으로 소개해 주세요.")
+
+print(response.text)
+print("Finish reason:", response.response_metadata.get("finish_reason"))
+print("Usage:", response.usage_metadata)

langchain_exaone-0.1.0/pyproject.toml

@@ -0,0 +1,42 @@
+[build-system]
+requires = ["hatchling>=1.26"]
+build-backend = "hatchling.build"
+
+[project]
+name = "langchain-exaone"
+version = "0.1.0"
+description = "LangChain chat model integration for the EXAONE API"
+readme = "README.md"
+requires-python = ">=3.10"
+license = "MIT"
+license-files = ["LICENSE"]
+dependencies = [
+    "httpx>=0.27,<1",
+    "langchain-core>=1.7.5,<2",
+    "pydantic>=2,<3",
+]
+
+[project.urls]
+Repository = "https://github.com/ykoh42/exaone-integrations"
+
+[dependency-groups]
+test = [
+    "langchain-tests==1.1.9",
+    "pytest>=9,<10",
+    "pytest-asyncio>=1,<2",
+]
+lint = ["ruff>=0.16,<0.18"]
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/langchain_exaone"]
+
+[tool.pytest.ini_options]
+asyncio_mode = "auto"
+asyncio_default_fixture_loop_scope = "function"
+testpaths = ["tests"]
+
+[tool.ruff]
+line-length = 120
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "B"]

langchain_exaone-0.1.0/src/langchain_exaone/__init__.py

@@ -0,0 +1,5 @@
+"""LangChain chat model integration for the EXAONE API."""
+
+from langchain_exaone.chat_models import ChatEXAONE, ExaoneAPIError
+
+__all__ = ["ChatEXAONE", "ExaoneAPIError"]