Merge pull request #52 from dontbanmeplz/main

feat(embedding): add AWS embedding provider

Author: nullure

.env.example

@@ -59,7 +59,7 @@ OM_WEAVIATE_CLASS=OpenMemory
 # --------------------------------------------
 # Embeddings Configuration
 # --------------------------------------------
-# Available providers: openai, gemini, ollama, local, synthetic
+# Available providers: openai, gemini, aws, ollama, local, synthetic
 # Embedding models per sector can be configured in models.yaml
 #
 # NOTE: Your selected TIER (fast/smart/deep) affects how embeddings work:
@@ -103,6 +103,11 @@ OPENAI_API_KEY=your-openai-api-key-here
 # Google Gemini Embeddings
 GEMINI_API_KEY=your-gemini-api-key-here
 
+# AWS Titan Text Embeddings V2
+AWS_ACCESS_KEY_ID=YOUR_ACCESS_KEY
+AWS_SECRET_ACCESS_KEY=YOUR_SECRET_KEY
+AWS_REGION=us-east-1
+
 # Ollama Local Embeddings
 OLLAMA_URL=http://localhost:11434
 

ARCHITECTURE.md

@@ -32,11 +32,11 @@ OpenMemory is a self-hosted AI memory engine implementing **Hierarchical Memory
 │               │         │                  │        │                  │
 │ • Classify    │         │ • OpenAI         │        │ • PDF Parser     │
 │ • Encode      │         │ • Gemini         │        │ • DOCX Parser    │
-│ • Store       │         │ • Ollama         │        │ • URL Scraper    │
-│ • Query       │         │ • Local/Synth    │        │ • Text Chunking  │
-│ • Decay       │         │ • Batch API      │        │                  │
-│ • Reinforce   │         └──────────────────┘        └──────────────────┘
-└───────┬───────┘
+│ • Store       │         │ • AWS            │        │ • URL Scraper    │
+│ • Query       │         │ • Ollama         │        │ • Text Chunking  │
+│ • Decay       │         │ • Local/Synth    │        │                  │
+│ • Reinforce   │         | • Batch API      |        └──────────────────┘
+└───────┬───────┘         └──────────────────┘
         │
         ├────────────────┐
         │                │
@@ -209,6 +209,7 @@ Memory A ──0.85──> Memory B
 | ------------- | -------------------------------------------------- | ------------- | ---------------- |
 | **OpenAI**    | `text-embedding-3-small`, `text-embedding-3-large` | ✅            | ~$0.02/1M tokens |
 | **Gemini**    | `embedding-001`                                    | ✅            | ~$0.01/1M tokens |
+| **AWS**       | `amazon.titan-embed-text-v2:0`                     | ❌            | ~$0.02/1M tokens |
 | **Ollama**    | `nomic-embed-text`, `bge-small`, `bge-large`       | ❌            | Free (local)     |
 | **Local**     | Custom models                                      | ❌            | Free (local)     |
 | **Synthetic** | Hash-based                                         | ❌            | Free             |
@@ -522,7 +523,7 @@ OM_DB_PATH=./data/openmemory.sqlite
 OM_API_KEY=                      # Optional bearer token
 
 # Embeddings
-OM_EMBEDDINGS=openai             # openai|gemini|ollama|local|synthetic
+OM_EMBEDDINGS=openai             # openai|gemini|aws|ollama|local|synthetic
 OM_EMBED_MODE=simple             # simple|advanced
 OM_ADV_EMBED_PARALLEL=false      # Parallel in advanced mode
 OM_EMBED_DELAY_MS=200            # Delay between calls
@@ -534,6 +535,10 @@ OPENAI_API_KEY=sk-...
 # Gemini
 GEMINI_API_KEY=...
 
+# AWS
+AWS_ACCESS_KEY_ID=...
+AWS_SECRET_ACCESS_KEY=...
+AWS_REGION="us-east-1"
 # Ollama
 OLLAMA_URL=http://localhost:11434
 

README.md

@@ -108,7 +108,7 @@ This approach improves recall accuracy while reducing costs.
 ### Summary
 
 OpenMemory delivers **2–3× faster contextual recall**, **6–10× lower cost**, and **full transparency** compared to hosted "memory APIs" like Zep or Supermemory.  
-Its **multi-sector cognitive model** allows explainable recall paths, hybrid embeddings (OpenAI / Gemini / Ollama / local), and real-time decay, making it ideal for developers seeking open, private, and interpretable long-term memory for LLMs.
+Its **multi-sector cognitive model** allows explainable recall paths, hybrid embeddings (OpenAI / Gemini / AWS / Ollama / local), and real-time decay, making it ideal for developers seeking open, private, and interpretable long-term memory for LLMs.
 
 ---
 
@@ -142,7 +142,7 @@ Requirements:
 
 - Node.js 20 or higher
 - SQLite 3.40 or higher (included)
-- Optional: OpenAI/Gemini API key or Ollama
+- Optional: OpenAI/Gemini/AWS API key or Ollama
 
 ```bash
 git clone https://github.com/caviraoss/openmemory.git
@@ -230,7 +230,7 @@ OpenMemory uses Hierarchical Memory Decomposition (HMD):
 
 - Backend: TypeScript
 - Storage: SQLite or PostgreSQL
-- Embeddings: E5/BGE/OpenAI/Gemini/Ollama
+- Embeddings: E5/BGE/OpenAI/Gemini/AWS/Ollama
 - Scheduler: node-cron for decay and maintenance
 
 **Query flow:**

Why.md

@@ -109,17 +109,17 @@ No vector DB can do this natively.
 
 ## 6. Developer Value Proposition
 
-| Area               | Why OpenMemory Wins                                   |
-| ------------------ | ----------------------------------------------------- |
-| **Setup**          | Single binary or Docker; no external service required |
-| **Integrations**   | Works with any LLM (OpenAI, Gemini, Ollama, Claude)   |
-| **SDKs**           | TypeScript + Python included                          |
-| **Cost**           | 10× cheaper than hosted alternatives                  |
-| **Explainability** | Memory formation and recall fully transparent         |
-| **Performance**    | p95 retrieval under 130ms on 100k+ nodes              |
-| **Scalability**    | Horizontally shardable by sector                      |
-| **Extensibility**  | Add new embedding models with 1 config line           |
-| **Privacy**        | Zero vendor lock-in; full data control                |
+| Area               | Why OpenMemory Wins                                      |
+| ------------------ | ---------------------------------------------------------|
+| **Setup**          | Single binary or Docker; no external service required    |
+| **Integrations**   | Works with any LLM (OpenAI, Gemini, AWS, Ollama, Claude) |
+| **SDKs**           | TypeScript + Python included                             |
+| **Cost**           | 10× cheaper than hosted alternatives                     |
+| **Explainability** | Memory formation and recall fully transparent            |
+| **Performance**    | p95 retrieval under 130ms on 100k+ nodes                 |
+| **Scalability**    | Horizontally shardable by sector                         |
+| **Extensibility**  | Add new embedding models with 1 config line              |
+| **Privacy**        | Zero vendor lock-in; full data control                   |
 
 ---