nylonme-integrations 0.1.0__tar.gz

nylonme_integrations-0.1.0/PKG-INFO

@@ -0,0 +1,88 @@
+Metadata-Version: 2.4
+Name: nylonme-integrations
+Version: 0.1.0
+Summary: LangChain and LlamaIndex adapters for the NylonME memory engine
+Author: NylonME contributors
+License: Apache-2.0
+Keywords: memory,rag,langchain,llamaindex,nylonme
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: nylon-sdk>=0.2
+Provides-Extra: langchain
+Requires-Dist: langchain-core>=0.3; extra == "langchain"
+Provides-Extra: llamaindex
+Requires-Dist: llama-index-core>=0.11; extra == "llamaindex"
+Provides-Extra: all
+Requires-Dist: langchain-core>=0.3; extra == "all"
+Requires-Dist: llama-index-core>=0.11; extra == "all"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+
+# NylonME Integrations
+
+LangChain 和 LlamaIndex 的 NylonME 适配器。核心共振逻辑只依赖
+`nylon-sdk`，框架类在真正使用时才导入，避免强制安装框架。
+
+## 安装
+
+```bash
+# LangChain
+pip install nylonme-integrations[langchain]
+
+# LlamaIndex
+pip install nylonme-integrations[llamaindex]
+
+# 两者都要
+pip install nylonme-integrations[all]
+```
+
+## LangChain
+
+```python
+from nylonme_integrations.langchain import NylonMeRetriever, NylonMeMemory
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+docs = retriever.invoke("上次出差住在哪里？")
+for d in docs:
+    print(d.page_content, d.metadata["resonance"])
+```
+
+对话记忆（在 prompt 里放一个 `{memory}` 占位符）：
+
+```python
+memory = NylonMeMemory(target="127.0.0.1:50051", owner="alice")
+memory.save_context({"input": "我喜欢靠窗座位"}, {"output": "已记住"})
+print(memory.load_memory_variables({"input": "出差选座"}))
+# {'memory': '- 出差偏好：靠窗座位'}
+```
+
+## LlamaIndex
+
+```python
+from nylonme_integrations.llamaindex import NylonMeRetriever
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+nodes = retriever.retrieve("上次出差住在哪里？")
+for n in nodes:
+    print(n.node.text, n.score)
+```
+
+## 配置
+
+两个适配器都暴露同样的字段：
+
+| 字段 | 默认值 | 说明 |
+|---|---|---|
+| `target` | `127.0.0.1:50051` | NylonME gRPC 地址（也读 `NYLON_SERVER`） |
+| `owner` | `default` | 记忆归属（也读 `NYLON_OWNER`） |
+| `tenant` | `default` | 租户（也读 `NYLON_TENANT`） |
+| `budget` | `5` | 共振召回预算 |
+| `task` | `None` | 可选任务上下文 |
+
+引擎默认监听 `127.0.0.1:50051`（gRPC）和 `50052`（REST/UI）。用 Docker 一键起：
+`docker compose up -d`。

nylonme_integrations-0.1.0/README.md

@@ -0,0 +1,63 @@
+# NylonME Integrations
+
+LangChain 和 LlamaIndex 的 NylonME 适配器。核心共振逻辑只依赖
+`nylon-sdk`，框架类在真正使用时才导入，避免强制安装框架。
+
+## 安装
+
+```bash
+# LangChain
+pip install nylonme-integrations[langchain]
+
+# LlamaIndex
+pip install nylonme-integrations[llamaindex]
+
+# 两者都要
+pip install nylonme-integrations[all]
+```
+
+## LangChain
+
+```python
+from nylonme_integrations.langchain import NylonMeRetriever, NylonMeMemory
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+docs = retriever.invoke("上次出差住在哪里？")
+for d in docs:
+    print(d.page_content, d.metadata["resonance"])
+```
+
+对话记忆（在 prompt 里放一个 `{memory}` 占位符）：
+
+```python
+memory = NylonMeMemory(target="127.0.0.1:50051", owner="alice")
+memory.save_context({"input": "我喜欢靠窗座位"}, {"output": "已记住"})
+print(memory.load_memory_variables({"input": "出差选座"}))
+# {'memory': '- 出差偏好：靠窗座位'}
+```
+
+## LlamaIndex
+
+```python
+from nylonme_integrations.llamaindex import NylonMeRetriever
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+nodes = retriever.retrieve("上次出差住在哪里？")
+for n in nodes:
+    print(n.node.text, n.score)
+```
+
+## 配置
+
+两个适配器都暴露同样的字段：
+
+| 字段 | 默认值 | 说明 |
+|---|---|---|
+| `target` | `127.0.0.1:50051` | NylonME gRPC 地址（也读 `NYLON_SERVER`） |
+| `owner` | `default` | 记忆归属（也读 `NYLON_OWNER`） |
+| `tenant` | `default` | 租户（也读 `NYLON_TENANT`） |
+| `budget` | `5` | 共振召回预算 |
+| `task` | `None` | 可选任务上下文 |
+
+引擎默认监听 `127.0.0.1:50051`（gRPC）和 `50052`（REST/UI）。用 Docker 一键起：
+`docker compose up -d`。

nylonme_integrations-0.1.0/pyproject.toml

@@ -0,0 +1,33 @@
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "nylonme-integrations"
+version = "0.1.0"
+description = "LangChain and LlamaIndex adapters for the NylonME memory engine"
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "Apache-2.0" }
+authors = [{ name = "NylonME contributors" }]
+keywords = ["memory", "rag", "langchain", "llamaindex", "nylonme"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = ["nylon-sdk>=0.2"]
+
+[project.optional-dependencies]
+langchain = ["langchain-core>=0.3"]
+llamaindex = ["llama-index-core>=0.11"]
+all = ["langchain-core>=0.3", "llama-index-core>=0.11"]
+test = ["pytest"]
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+nylonme_integrations = ["py.typed"]

nylonme_integrations-0.1.0/setup.cfg

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

nylonme_integrations-0.1.0/src/nylonme_integrations/__init__.py

@@ -0,0 +1,18 @@
+"""NylonME framework adapters.
+
+A thin bridge from NylonME to the two most common Python agent/RAG
+frameworks. The core resonance logic lives in ``_core.py`` and only depends
+on ``nylon-sdk``; the framework modules import their framework lazily.
+
+    # LangChain
+    from nylonme_integrations.langchain import NylonMeRetriever, NylonMeMemory
+
+    # LlamaIndex
+    from nylonme_integrations.llamaindex import NylonMeRetriever
+"""
+
+from ._core import build_client, node_metadata, resonate_records
+
+__version__ = "0.1.0"
+
+__all__ = ["build_client", "node_metadata", "resonate_records", "__version__"]

nylonme_integrations-0.1.0/src/nylonme_integrations/_core.py

@@ -0,0 +1,63 @@
+"""Framework-agnostic bridge between NylonME and RAG/agent frameworks.
+
+This module only depends on ``nylon-sdk``. The LangChain and LlamaIndex
+adapters import it, so the same resonate/weave semantics are shared and can
+be tested without pulling in either framework.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Optional
+
+from nylon_sdk import NylonClient
+
+
+def build_client(
+    target: str,
+    owner: str,
+    tenant: str,
+    *,
+    timeout: float = 30.0,
+) -> NylonClient:
+    """Create a NylonME client from the values the adapters expose as fields."""
+    return NylonClient(target, owner=owner, tenant=tenant, timeout=timeout)
+
+
+def node_metadata(record: dict[str, Any]) -> dict[str, Any]:
+    """Flatten one ActivatedNode into stable, JSON-friendly metadata."""
+    f = record["filaments"]
+    return {
+        "node_id": record["node_id"],
+        "resonance": record["resonance"],
+        "relations": f["relations"],
+        "confidence": f["confidence"],
+        "emotion_valence": f["emotion_valence"],
+        "emotion_intensity": f["emotion_intensity"],
+        "mentions_7d": f["mentions_7d"],
+    }
+
+
+def resonate_records(
+    client: NylonClient,
+    query: str,
+    *,
+    budget: int = 5,
+    task: Optional[str] = None,
+) -> list[dict[str, Any]]:
+    """Return resonated memories as plain dictionaries."""
+    result = client.resonate(query, budget=budget, task=task)
+    return [
+        {
+            "node_id": a.node_id,
+            "resonance": a.resonance,
+            "filaments": {
+                "fact": a.filaments.fact,
+                "relations": list(a.filaments.relations),
+                "confidence": a.filaments.confidence,
+                "emotion_valence": a.filaments.emotion_valence,
+                "emotion_intensity": a.filaments.emotion_intensity,
+                "mentions_7d": a.filaments.mentions_7d,
+            },
+        }
+        for a in result.activated
+    ]

nylonme_integrations-0.1.0/src/nylonme_integrations/langchain.py

@@ -0,0 +1,114 @@
+"""LangChain adapter for NylonME.
+
+Exposes:
+
+* :class:`NylonMeRetriever` - RAG retrieval over resonated memories.
+* :class:`NylonMeMemory` - conversational memory that weaves turns and
+  injects the most relevant memories into the prompt.
+
+``langchain-core`` is a required dependency of this module; install it with
+``pip install nylonme-integrations[langchain]``.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Optional
+
+try:
+    from langchain_core.callbacks.manager import CallbackManagerForRetrieverRun
+    from langchain_core.documents import Document
+    from langchain_core.memory import BaseMemory
+    from langchain_core.retrievers import BaseRetriever
+except ImportError as exc:  # pragma: no cover
+    raise ImportError(
+        "This adapter requires langchain-core. "
+        "Install it with: pip install nylonme-integrations[langchain]"
+    ) from exc
+
+from ._core import build_client, node_metadata, resonate_records
+
+
+class NylonMeRetriever(BaseRetriever):
+    """Retriever that maps a query to resonated NylonME memories as Documents."""
+
+    target: str = "127.0.0.1:50051"
+    owner: str = "default"
+    tenant: str = "default"
+    budget: int = 5
+    task: Optional[str] = None
+
+    def _get_relevant_documents(
+        self,
+        query: str,
+        *,
+        run_manager: CallbackManagerForRetrieverRun,
+    ) -> list[Document]:
+        del run_manager
+        with build_client(self.target, self.owner, self.tenant) as client:
+            records = resonate_records(client, query, budget=self.budget, task=self.task)
+        return [
+            Document(page_content=r["filaments"]["fact"], metadata=node_metadata(r))
+            for r in records
+        ]
+
+    async def _aget_relevant_documents(
+        self,
+        query: str,
+        *,
+        run_manager=None,
+    ) -> list[Document]:
+        del run_manager
+        # The sync SDK call is the simplest correct path; async callers can
+        # use AsyncNylonClient directly if they need a truly async channel.
+        return self._get_relevant_documents(query, run_manager=run_manager)
+
+
+class NylonMeMemory(BaseMemory):
+    """Conversational memory: weaves each turn, injects recall into prompts.
+
+    Wire it into a prompt that contains a ``{memory}`` placeholder (or the
+    ``memory_key`` you configure). On ``save_context`` the latest user input
+    and model output are woven into the engine; on ``load_memory_variables``
+    the engine resonates on the current input and returns the matched facts.
+    """
+
+    target: str = "127.0.0.1:50051"
+    owner: str = "default"
+    tenant: str = "default"
+    budget: int = 4
+    memory_key: str = "memory"
+    input_key: str = "input"
+
+    @property
+    def memory_variables(self) -> list[str]:
+        return [self.memory_key]
+
+    def load_memory_variables(self, inputs: dict[str, Any]) -> dict[str, str]:
+        query = self._query_from(inputs)
+        if not query:
+            return {self.memory_key: ""}
+        with build_client(self.target, self.owner, self.tenant) as client:
+            records = resonate_records(client, query, budget=self.budget)
+        facts = [r["filaments"]["fact"] for r in records]
+        return {self.memory_key: "\n".join(f"- {f}" for f in facts)}
+
+    def save_context(self, inputs: dict[str, Any], outputs: dict[str, str]) -> None:
+        text = " ".join(
+            str(v) for v in list(inputs.values()) + list(outputs.values()) if v
+        ).strip()
+        if text:
+            with build_client(self.target, self.owner, self.tenant) as client:
+                client.weave(text)
+
+    def clear(self) -> None:
+        # The community engine has no destructive clear; memories decay via
+        # the tension model rather than being force-deleted.
+        return None
+
+    def _query_from(self, inputs: dict[str, Any]) -> str:
+        if self.input_key in inputs:
+            return str(inputs[self.input_key])
+        return " ".join(str(v) for v in inputs.values() if v).strip()
+
+
+__all__ = ["NylonMeRetriever", "NylonMeMemory"]

nylonme_integrations-0.1.0/src/nylonme_integrations/llamaindex.py

@@ -0,0 +1,69 @@
+"""LlamaIndex adapter for NylonME.
+
+Exposes :class:`NylonMeRetriever`, which wraps NylonME resonance recall as a
+LlamaIndex retriever returning ``NodeWithScore`` objects.
+
+``llama-index-core`` is a required dependency of this module; install it with
+``pip install nylonme-integrations[llamaindex]``.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Optional
+
+try:
+    from llama_index.core.retrievers import BaseRetriever
+    from llama_index.core.schema import NodeWithScore, QueryBundle, TextNode
+except ImportError as exc:  # pragma: no cover
+    raise ImportError(
+        "This adapter requires llama-index-core. "
+        "Install it with: pip install nylonme-integrations[llamaindex]"
+    ) from exc
+
+from ._core import build_client, node_metadata, resonate_records
+
+
+class NylonMeRetriever(BaseRetriever):
+    """LlamaIndex retriever backed by NylonME resonance recall."""
+
+    def __init__(
+        self,
+        target: str = "127.0.0.1:50051",
+        owner: str = "default",
+        tenant: str = "default",
+        budget: int = 5,
+        task: Optional[str] = None,
+        **kwargs: Any,
+    ) -> None:
+        # llama-index-core >= 0.11 BaseRetriever is a plain class with an
+        # explicit __init__ (no pydantic fields), so config must be stored
+        # as instance attributes after super().__init__.
+        super().__init__(**kwargs)
+        self.target = target
+        self.owner = owner
+        self.tenant = tenant
+        self.budget = budget
+        self.task = task
+
+    def _retrieve(self, query_bundle: QueryBundle) -> list[NodeWithScore]:
+        query = query_bundle.query_str
+        with build_client(self.target, self.owner, self.tenant) as client:
+            records = resonate_records(client, query, budget=self.budget, task=self.task)
+        return [
+            NodeWithScore(
+                node=TextNode(
+                    text=r["filaments"]["fact"],
+                    id_=str(r["node_id"]),
+                    metadata=node_metadata(r),
+                ),
+                score=float(r["resonance"]),
+            )
+            for r in records
+        ]
+
+    async def _aretrieve(self, query_bundle: QueryBundle) -> list[NodeWithScore]:
+        # Sync SDK call on the event loop for a simple, correct default.
+        return self._retrieve(query_bundle)
+
+
+__all__ = ["NylonMeRetriever"]

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

@@ -0,0 +1,88 @@
+Metadata-Version: 2.4
+Name: nylonme-integrations
+Version: 0.1.0
+Summary: LangChain and LlamaIndex adapters for the NylonME memory engine
+Author: NylonME contributors
+License: Apache-2.0
+Keywords: memory,rag,langchain,llamaindex,nylonme
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: nylon-sdk>=0.2
+Provides-Extra: langchain
+Requires-Dist: langchain-core>=0.3; extra == "langchain"
+Provides-Extra: llamaindex
+Requires-Dist: llama-index-core>=0.11; extra == "llamaindex"
+Provides-Extra: all
+Requires-Dist: langchain-core>=0.3; extra == "all"
+Requires-Dist: llama-index-core>=0.11; extra == "all"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+
+# NylonME Integrations
+
+LangChain 和 LlamaIndex 的 NylonME 适配器。核心共振逻辑只依赖
+`nylon-sdk`，框架类在真正使用时才导入，避免强制安装框架。
+
+## 安装
+
+```bash
+# LangChain
+pip install nylonme-integrations[langchain]
+
+# LlamaIndex
+pip install nylonme-integrations[llamaindex]
+
+# 两者都要
+pip install nylonme-integrations[all]
+```
+
+## LangChain
+
+```python
+from nylonme_integrations.langchain import NylonMeRetriever, NylonMeMemory
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+docs = retriever.invoke("上次出差住在哪里？")
+for d in docs:
+    print(d.page_content, d.metadata["resonance"])
+```
+
+对话记忆（在 prompt 里放一个 `{memory}` 占位符）：
+
+```python
+memory = NylonMeMemory(target="127.0.0.1:50051", owner="alice")
+memory.save_context({"input": "我喜欢靠窗座位"}, {"output": "已记住"})
+print(memory.load_memory_variables({"input": "出差选座"}))
+# {'memory': '- 出差偏好：靠窗座位'}
+```
+
+## LlamaIndex
+
+```python
+from nylonme_integrations.llamaindex import NylonMeRetriever
+
+retriever = NylonMeRetriever(target="127.0.0.1:50051", owner="alice", budget=5)
+nodes = retriever.retrieve("上次出差住在哪里？")
+for n in nodes:
+    print(n.node.text, n.score)
+```
+
+## 配置
+
+两个适配器都暴露同样的字段：
+
+| 字段 | 默认值 | 说明 |
+|---|---|---|
+| `target` | `127.0.0.1:50051` | NylonME gRPC 地址（也读 `NYLON_SERVER`） |
+| `owner` | `default` | 记忆归属（也读 `NYLON_OWNER`） |
+| `tenant` | `default` | 租户（也读 `NYLON_TENANT`） |
+| `budget` | `5` | 共振召回预算 |
+| `task` | `None` | 可选任务上下文 |
+
+引擎默认监听 `127.0.0.1:50051`（gRPC）和 `50052`（REST/UI）。用 Docker 一键起：
+`docker compose up -d`。

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

@@ -0,0 +1,12 @@
+README.md
+pyproject.toml
+src/nylonme_integrations/__init__.py
+src/nylonme_integrations/_core.py
+src/nylonme_integrations/langchain.py
+src/nylonme_integrations/llamaindex.py
+src/nylonme_integrations/py.typed
+src/nylonme_integrations.egg-info/PKG-INFO
+src/nylonme_integrations.egg-info/SOURCES.txt
+src/nylonme_integrations.egg-info/dependency_links.txt
+src/nylonme_integrations.egg-info/requires.txt
+src/nylonme_integrations.egg-info/top_level.txt

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

@@ -0,0 +1 @@
+

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

@@ -0,0 +1,14 @@
+nylon-sdk>=0.2
+
+[all]
+langchain-core>=0.3
+llama-index-core>=0.11
+
+[langchain]
+langchain-core>=0.3
+
+[llamaindex]
+llama-index-core>=0.11
+
+[test]
+pytest

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

@@ -0,0 +1 @@
+nylonme_integrations