Migrates the current redis package from NAT repo to this repo

README.md

@@ -1,15 +1,50 @@
-# nvidia-nat-redis
+<div align="center">
+<img src="assets/redis-logo.svg" alt="Redis" width="175px">
 
-Redis Agent Memory integrations for
-[NVIDIA NeMo Agent Toolkit](https://github.com/NVIDIA/NeMo-Agent-Toolkit).
+# Nvidia-NAT-Redis
+</div>
 
-This standalone plugin exposes two NAT surfaces:
+**Redis-backed memory for [NVIDIA NeMo Agent Toolkit](https://github.com/NVIDIA/NeMo-Agent-Toolkit)** — production [Redis Agent Memory](https://redis.github.io/agent-memory-server/quick-start/) integrations plus direct Redis plugins (the historical `nat.plugins.redis` stack).
 
-1. `_type: redis_agent_memory_backend`
-   Redis Agent Memory as a NAT `MemoryEditor` long-term memory backend.
-2. `_type: redis_agent_memory_auto_memory`
-   A native Redis Agent Memory wrapper that uses working memory plus
-   `memory_prompt` hydration on every turn.
+In NAT, long-term recall is usually wired through memory tools such as `get_memory` and `add_memory`, which delegate to a **`MemoryEditor`** implementation. This package adds Redis-backed options in **two families**: **Redis Agent Memory** (AMS over HTTP — richer lifecycle, optional auto-memory workflow) and **direct Redis** (JSON + vector search or plain KV inside `nat.plugins.redis`, no AMS).
+
+This repo is the standalone home for integrations that used to ship under
+[`packages/nvidia_nat_redis`](https://github.com/NVIDIA/NeMo-Agent-Toolkit/tree/develop/packages/nvidia_nat_redis) in the NeMo Agent Toolkit monorepo.
+
+### Redis Agent Memory (full stack)
+1. `_type: redis_agent_memory_backend` — Redis Agent Memory as a NAT `MemoryEditor` long-term memory backend.
+2. `_type: redis_agent_memory_auto_memory` — A native Redis Agent Memory wrapper that uses working memory plus
+`memory_prompt` hydration on every turn.
+AMS runs as a separate service; these surfaces talk to it via the agent-memory client.
+
+1. `_type: redis_agent_memory_backend` — Redis Agent Memory behind NAT’s standard **`MemoryEditor`** contract (fits tool-driven long-term memory).
+2. `_type: redis_agent_memory_auto_memory` — A **`workflow` wrapper** (not just another editor): working memory, `memory_prompt` hydration every turn, turn capture, and promotion — the fullest AMS-shaped integration.
+
+### Direct Redis (simple in-Redis memory)
+
+Loaded via the `nat_redis` setuptools entry point (same name as the historical
+monorepo package):
+
+- `_type: redis_memory` — Vector search over JSON documents in Redis (RediSearch); requires a workflow `embedder` and Redis Stack (or Redis with search + JSON support). Implements **`MemoryEditor`**-style semantic memory without AMS.
+- `_type: redis` — NAT **object store** on plain Redis key–value storage (not semantic long-term memory).
+
+| | **Redis Agent Memory** | **Direct Redis** |
+| --- | --- | --- |
+| **Runs** | AMS service + Redis | Redis only (your NAT process uses the client) |
+| **Best for** | Learning-style memory, working memory, AMS filters and APIs | Lightweight Redis-native memory or KV object store |
+| **Tradeoff** | Operate AMS; HTTP path | Simpler ops for `redis` / `redis_memory`; vector path needs embedder + search-capable Redis |
+
+Use **Redis Agent Memory** when you want the Redis Agent Memory feature set.
+Use **`redis_memory`** when you only need a lightweight Redis-native `MemoryEditor`
+without AMS.
+
+### Python imports (same as NeMo)
+
+Direct Redis support is implemented only under **`nat.plugins.redis`** (for example
+`from nat.plugins.redis.redis_editor import RedisEditor`), matching NeMo Agent
+Toolkit. The setuptools entry point **`nat_redis`** loads
+`nat.plugins.redis.register`. Redis Agent Memory code lives under
+**`nvidia_nat_redis.redis_agent_memory`**.
 
 
 ## Install
@@ -30,9 +65,13 @@ uv sync --group dev --extra test
 
 - Use `_type: redis_agent_memory_backend` when your workflow already uses NAT memory tools and you want Redis Agent Memory behind the standard `MemoryEditor` contract.
 - Use `_type: redis_agent_memory_auto_memory` when you want Redis Agent Memory to own working-memory continuity, prompt hydration, and turn capture on every request. This exposes the richness of Redis Agent Memory in its fullest form.
+- Use `_type: redis_memory` when you want the simpler direct-Redis memory from NeMo Agent Toolkit (Redis JSON + vector index, no Redis Agent Memory). You must configure an `embedder` reference and run a Redis deployment that supports the search commands used by the plugin.
+- Use `_type: redis` when you need NAT’s Redis-backed **object store** (KV), not vector or AMS-backed semantic memory.
 
 ## Integration Modes
 
+These examples cover the two **Redis Agent Memory** integration shapes (`MemoryEditor` backend vs `workflow` wrapper). For **`redis_memory`** / **`redis`**, see [Configuration reference](docs/configuration.md).
+
 ### Direct Long-Term Memory Backend
 
 Use `_type: redis_agent_memory_backend` when you want Redis Agent Memory behind NAT's

docs/configuration.md

@@ -1,9 +1,25 @@
 # Configuration
 
-This package ships two Redis Agent Memory surfaces for NAT:
+This package ships Redis Agent Memory surfaces and direct Redis plugins (migrated
+from the NeMo Agent Toolkit `nvidia_nat_redis` package).
 
-1. A long-term memory backend: `_type: redis_agent_memory_backend`
-2. A native automatic wrapper: `_type: redis_agent_memory_auto_memory`
+Redis Agent Memory is a production-ready agent memory layer that extracts and stores
+relevant information and learns over time.
+
+Direct Redis memory provides a more standard semantic memory layer for LLM applications.
+
+**Full Redis Agent Memory auto wrapper**
+
+1. Long-term memory backend: `_type: redis_agent_memory_backend`
+2. Native automatic wrapper: `_type: redis_agent_memory_auto_memory`
+
+**Direct Redis (entry point `nat_redis` → `nat.plugins.redis.register`)**
+
+3. In-Redis vector memory: `_type: redis_memory`
+4. Object store: `_type: redis`
+
+For Python code, use **`nat.plugins.redis.*`** — the same module layout as NeMo
+Agent Toolkit’s in-tree package.
 
 ## Long-Term Memory Backend
 
@@ -116,6 +132,50 @@ Wrapper flow:
 2. Create or load Redis Agent Memory working memory for that identity.
 3. Call `memory_prompt(...)`, invoke the inner chat function with the hydrated request, and append the finished turn back into working memory.
 
+## Direct Redis memory (`redis_memory`)
+
+Use `_type: redis_memory` for the lightweight Redis-backed `MemoryEditor` that
+stores JSON documents and uses RediSearch vector queries. This matches the
+behavior of the standalone
+[`nvidia_nat_redis` package in NeMo Agent Toolkit](https://github.com/NVIDIA/NeMo-Agent-Toolkit/tree/develop/packages/nvidia_nat_redis).
+
+Requirements:
+
+- Redis with **JSON** and **search** (for example Redis Stack), because the implementation creates a JSON index and runs vector KNN queries.
+- A workflow **embedder** instance; `embedder` must reference its name in the workflow configuration.
+
+```yaml
+memory:
+  redis_ltm:
+    _type: redis_memory
+    host: localhost
+    port: 6379
+    db: 0
+    key_prefix: nat
+    embedder: my_openai_embedder
+```
+
+Supported config fields:
+
+- `host`, `port`, `db`, `password` (optional secret), `key_prefix`
+- `embedder`: `EmbedderRef` to a configured embedder (LangChain wrapper)
+
+## Direct Redis object store (`redis`)
+
+Use `_type: redis` for the NAT object store that persists `ObjectStoreItem`
+JSON at keys under `nat/object_store/{bucket_name}/...`, with optional TTL.
+
+```yaml
+object_store:
+  artifacts:
+    _type: redis
+    host: localhost
+    port: 6379
+    db: 0
+    bucket_name: my_bucket
+    ttl: 3600
+```
+
 ## Examples
 
 - [Redis Agent Memory native wrapper](../examples/agent_auto_memory/README.md)

examples/agent_auto_memory/README.md

@@ -40,7 +40,7 @@ The example expects:
 
 ## Start Services
 
-Compose starts Redis Stack and Agent Memory Server for local development. Both
+Compose starts Redis Stack and Redis Agent Memory containers for local development. Both
 ports bind to `127.0.0.1`, and AMS auth is disabled.
 
 ```bash

pyproject.toml

@@ -21,7 +21,9 @@ classifiers = [
 ]
 dependencies = [
   "agent-memory-client>=0.14.0,<1.0.0",
+  "langchain-core>=0.3.0,<2.0.0",
   "nvidia-nat-core>=1.5.0,<2.0.0",
+  "redis>=4.3.4,<5.0.0",
 ]
 
 [project.optional-dependencies]
@@ -38,6 +40,7 @@ source = "https://github.com/redis-developer/nvidia-nat-redis"
 
 [project.entry-points.'nat.components']
 redis_agent_memory = "nvidia_nat_redis.redis_agent_memory.register"
+nat_redis = "nat.plugins.redis.register"
 
 [dependency-groups]
 dev = [
@@ -47,7 +50,7 @@ dev = [
 
 [tool.setuptools.packages.find]
 where = ["src"]
-include = ["nvidia_nat_redis*"]
+include = ["nvidia_nat_redis*", "nat.plugins*"]
 
 [tool.setuptools_scm]
 git_describe_command = "git describe --long --first-parent"

src/nat/plugins/redis/__init__.py

@@ -0,0 +1,16 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+"""Direct Redis memory and object-store NAT components (NeMo ``nat.plugins.redis`` layout)."""

src/nat/plugins/redis/memory.py

@@ -0,0 +1,60 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+from nat.builder.builder import Builder
+from nat.cli.register_workflow import register_memory
+from nat.data_models.common import OptionalSecretStr, get_secret_value
+from nat.data_models.component_ref import EmbedderRef
+from nat.data_models.memory import MemoryBaseConfig
+from pydantic import Field
+
+
+class RedisMemoryClientConfig(MemoryBaseConfig, name="redis_memory"):
+    host: str = Field(default="localhost", description="Redis server host")
+    db: int = Field(default=0, description="Redis DB")
+    port: int = Field(default=6379, description="Redis server port")
+    password: OptionalSecretStr = Field(default=None, description="Password for the Redis server")
+    key_prefix: str = Field(default="nat", description="Key prefix to use for redis keys")
+    embedder: EmbedderRef = Field(description=("Instance name of the memory client instance from the workflow "
+                                               "configuration object."))
+
+
+@register_memory(config_type=RedisMemoryClientConfig)
+async def redis_memory_client(config: RedisMemoryClientConfig, builder: Builder):
+
+    from nat.builder.framework_enum import LLMFrameworkEnum
+
+    import redis.asyncio as redis
+
+    from .redis_editor import RedisEditor
+    from .schema import ensure_index_exists
+
+    redis_client = redis.Redis(host=config.host,
+                               port=config.port,
+                               db=config.db,
+                               password=get_secret_value(config.password),
+                               decode_responses=True,
+                               socket_timeout=5.0,
+                               socket_connect_timeout=5.0)
+
+    embedder = await builder.get_embedder(config.embedder, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
+
+    test_embedding = await embedder.aembed_query("test")
+    embedding_dim = len(test_embedding)
+    await ensure_index_exists(client=redis_client, key_prefix=config.key_prefix, embedding_dim=embedding_dim)
+
+    memory_editor = RedisEditor(redis_client=redis_client, key_prefix=config.key_prefix, embedder=embedder)
+
+    yield memory_editor