feat: add AgentsKit Atlas vector memory example

apps/README.md

@@ -2,6 +2,7 @@ Javascript and Python apps and demos showcasing how to use MongoDB in GenAI appl
 
 | Application | Stack | Description | Directory |
 |------------|-------|-------------|-----------|
+| AgentsKit Atlas Memory | MongoDB Atlas, AgentsKit, TypeScript | Minimal vector-memory lifecycle with replaceable embedding and model providers | [![View Code](https://img.shields.io/badge/view-code-blue?logo=github)](agentskit-atlas-memory) |
 | Ambient Inventory Agent | MongoDB Atlas, MongoDB Remote MCP, Python, FastAPI, LangGraph, Claude on Amazon Bedrock | Coffee roaster inventory assistant: a deterministic LangGraph monitor raises low-stock alerts, and a streaming LangGraph ReAct agent answers the owner's questions using MongoDB Remote MCP tools | [![View Code](https://img.shields.io/badge/view-code-blue?logo=github)](ambient-inventory-agent) |
 | Graph RAG Demo | MongoDB Atlas, Python, Node.js, OpenAI | Graph-based RAG implementation demonstrating MongoDB's graph capabilities with hybrid Python/Node.js approach | [![View Code](https://img.shields.io/badge/view-code-blue?logo=github)](graph_rag_demo) |
 | Local RAG PDF | MongoDB Atlas, Python | Local RAG implementation for PDF processing and querying | [![View Code](https://img.shields.io/badge/view-code-blue?logo=github)](local-rag-pdf) |

apps/agentskit-atlas-memory/.env.example

@@ -0,0 +1,4 @@
+MONGODB_URI=
+MONGODB_DATABASE=agentskit_demo
+MONGODB_COLLECTION=memory
+MONGODB_VECTOR_INDEX=agentskit_vector_index

apps/agentskit-atlas-memory/.gitignore

@@ -0,0 +1,3 @@
+.env
+node_modules/
+dist/

apps/agentskit-atlas-memory/README.md

@@ -0,0 +1,60 @@
+# AgentsKit memory with MongoDB Atlas Vector Search
+
+This minimal TypeScript example uses MongoDB Atlas as a replaceable vector-memory
+backend for an agent. It exercises the complete memory lifecycle: store three
+documents, retrieve the closest context, print the result, and remove the demo
+documents.
+
+The sample uses small deterministic vectors so the integration is easy to
+inspect. In an application, replace them with vectors from any embedding
+provider; the MongoDB collection and AgentsKit memory contract stay unchanged.
+
+## Prerequisites
+
+- Node.js 20 or newer
+- A MongoDB Atlas cluster
+- An Atlas Vector Search index named `agentskit_vector_index` on
+  `agentskit_demo.memory`
+
+Use this index definition for the sample's three-dimensional vectors:
+
+```json
+{
+  "fields": [
+    {
+      "type": "vector",
+      "path": "embedding",
+      "numDimensions": 3,
+      "similarity": "cosine"
+    }
+  ]
+}
+```
+
+## Run the example
+
+```bash
+cp .env.example .env
+npm install
+node --env-file=.env --import tsx src/index.ts
+```
+
+Set `MONGODB_URI` in `.env` before running. The database, collection, and index
+names can also be changed there.
+
+Atlas Vector Search indexes new records asynchronously. The example retries the
+search at one-second intervals for up to five attempts while the stored records
+become queryable. It
+deletes its sample records before and after the run, including when indexing
+does not complete within that retry window, so repeated executions do not leave
+demo data behind.
+
+## Validate without credentials
+
+```bash
+npm run typecheck
+npm test
+```
+
+The deterministic test exercises the collection boundary and verifies the
+generated `$vectorSearch` pipeline without contacting Atlas.