Update README.md

README.md

@@ -13,52 +13,22 @@ tags:
   - agent-memory
   - semantic-search
   - neural-links
-  - context-management
-  - conversation-memory
-  - llm-memory
-  - ai-memory
   - mcp
 short_description: AI agent memory with neural links and smart injection
 ---
 
-# 🧠 Mnemo v4 — AI Agent Memory System
+# Mnemo v5.0 - AI Agent Memory System
 
-**MCP-enabled memory server for AI agents.** Store, search, and retrieve memories with real semantic embeddings, bidirectional neural links, smart injection decisions, and automatic decay/pruning.
+MCP-enabled memory server for AI agents with real semantic embeddings, bidirectional neural links, smart injection, and automatic decay/pruning.
 
-## What It Does
+## v5 Changes
 
-Mnemo gives AI agents persistent memory across conversations. Add facts, preferences, analysis results — then retrieve the most relevant context when needed. The smart injection predictor decides *when* memory helps vs. hurts, avoiding unnecessary context pollution.
+- Shared Core - single mnemo_core.py powers both spaces
+- Optional FAISS - O(log n) nearest neighbor search
+- Link Decay - links lose strength over time
+- Smart Injection v2 - cosine confidence replaces keyword-only heuristic
+- Tier Management - auto-promote to WORKING / demote to ARCHIVE
+- Embedding Persistence - base64 embeddings skip re-encoding on restart
+- Configurable - all weights via MnemoConfig dataclass
 
-## Features
-
-- **🔍 True Semantic Memory Search** — powered by local `all-MiniLM-L6-v2` embeddings to understand context and synonyms, enhanced with neural link boosting.
-- **🔗 Bidirectional Neural Links** — 8 link types connect related memories automatically based on semantic similarity and co-occurrence.
-- **🎯 Smart Injection Predictor** — decides when to inject memory vs. skip (avoids hurting in-conversation performance).
-- **🧹 Decay & Pruning** — unused memories degrade over time; stale ones get auto-removed to keep the memory space clean.
-- **🔌 MCP-Enabled** — works as an MCP tool in Claude Desktop, Cursor, VS Code, and other MCP clients.
-
-## Use Cases
-
-- Agent memory persistence across sessions
-- Knowledge base for chatbots and assistants
-- Context management for creative writing tools
-- Cross-conversation reference and recall
-
-## MCP Tools
-
-When added as an MCP server, Mnemo exposes these tools to your AI agent:
-
-| Tool | Description |
-|------|-------------|
-| `add_memory` | Store a new memory with optional namespace and metadata |
-| `search_memories` | Search memories by semantic similarity |
-| `check_injection` | Decide whether memory context would help the current query |
-| `get_context_for_injection` | Get formatted context block ready for prompt injection |
-| `delete_memory_by_id` | Remove a specific memory and its links |
-| `run_maintenance` | Trigger decay and pruning cycle |
-
-## Quick Start
-
-Add this Space as an MCP tool from the MCP badge above, or use the interactive demo tabs to try it manually.
-
-[MCP Server (REST API)](https://huggingface.co/spaces/AthelaPerk/mnemo-mcp) | [Library](https://huggingface.co/AthelaPerk/mnemo-memory) | MIT License
+[MCP Server](https://huggingface.co/spaces/AthelaPerk/mnemo-mcp) | MIT License