refactor: Clean up stale references (FAISS, ChromaDB, LanceDB)

H4LFdotDEV · H4LFdotDEV · commit 579ba0cedc58 · 2026-02-01T12:03:07.000-06:00
- Remove FAISS from requirements.txt (already removed from pyproject.toml)
- Update config/mcp-servers.json description to reflect Graphiti
- Replace FAISS config with Graphiti in settings.yaml
- Update Bruno API docs to remove FAISS references
- Fix test mocks to use graphiti instead of faiss
- Update CLAUDE.md architecture (ChromaDB → Neo4j)
- Update README.md search system (remove Hound, update roadmap)
- Fix wiki docs (Installation, Configuration, Troubleshooting)
- Register pytest 'slow' marker in pyproject.toml
- Remove lancedb directory creation from install.sh
- Add security contact email to SECURITY.md
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -17,7 +17,7 @@ An AI-native development environment that extends Claude Code with persistent me
 ├─────────────────────────────────────────────────────────────┤
 │  Infrastructure                                              │
 │  ├── Redis - Hot memory cache (port 6379)                   │
-│  ├── ChromaDB - Vector embeddings (port 8000)               │
+│  ├── Neo4j - Knowledge graph (port 7687)                    │
 │  └── LiteLLM - Model routing (port 4000)                    │
 └─────────────────────────────────────────────────────────────┘
 ```
@@ -241,7 +241,7 @@ docker-compose -f docker/docker-compose.yaml --profile local-llm up -d
 | Service | Port | Purpose |
 |---------|------|---------|
 | redis | 6379 | Hot memory cache |
-| chromadb | 8000 | Vector embeddings |
+| neo4j | 7474/7687 | Knowledge graph (HTTP/Bolt) |
 | litellm | 4000 | Model routing |
 | playwright | 9222 | Browser automation (optional) |
 | ollama | 11434 | Local LLM (optional) |
diff --git a/README.md b/README.md
@@ -6,8 +6,8 @@ An AI-native development environment that extends Claude Code with persistent me
 
 Claude Code++ adds enterprise-grade capabilities to Claude Code through MCP (Model Context Protocol) servers:
 
-- **Memory MCP** - Four-tier persistent memory (Redis → Graphiti → livegrep → Obsidian)
-- **Search MCP** - Multi-layer search (Hound → livegrep → Graphiti → Semantic)
+- **Memory MCP** - Four-tier persistent memory (Redis → Graphiti → SQLite → Obsidian)
+- **Search MCP** - Multi-layer search (SQLite FTS → livegrep → Graphiti)
 - **System Controller** - macOS Accessibility API integration
 - **Research Environment** - Voice conversations + webcam whiteboard capture
 - **Infrastructure** - Docker-based services for Redis, Neo4j, and model routing
@@ -243,10 +243,10 @@ See [wiki/Research-Environment.md](wiki/Research-Environment.md) for full docume
 
 ## Roadmap
 
-- [ ] Hound integration for project-local search
-- [ ] Semantic search layer (Nomic Embed + LanceDB)
-- [ ] Search MCP server (separate from Memory MCP)
+- [ ] Semantic search layer (intent-based queries)
 - [ ] Windows/Linux System Controller
+- [ ] CAIIDE++ VS Code fork integration
+- [ ] Voice mode improvements
 
 ## License
 
diff --git a/SECURITY.md b/SECURITY.md
@@ -8,6 +8,10 @@ If you discover a security vulnerability in Claude Code++, please report it resp
 2. Email security concerns to the maintainers directly
 3. Include details about the vulnerability and steps to reproduce
 
+## Security Contact
+
+Email: jeremiahk@halfservers.com
+
 ## Security Measures
 
 Claude Code++ implements several security measures:
diff --git a/bruno/memory-mcp/Memory Stats.bru b/bruno/memory-mcp/Memory Stats.bru
@@ -51,14 +51,14 @@ docs {
     - `sqlite`: SQLite cold storage
     - `vault`: Obsidian vault archive
     - `redis`: Redis hot cache
-    - `faiss`: FAISS vector search
+    - `graphiti`: Graphiti/Neo4j knowledge graph
     - `embedder`: Embedding provider
   - `health`: Detailed health status with latency
     - Each component shows: status, latency_ms, error (if any)
   - `sqlite`: SQLite-specific stats
   - `vault`: Vault-specific stats
   - `redis`: Redis-specific stats (if available)
-  - `faiss`: FAISS-specific stats (if available)
+  - `graphiti`: Graphiti-specific stats (if available)
   - `embedder`: Embedder info (if available)
 
   ## Use Cases
diff --git a/bruno/memory-mcp/README.md b/bruno/memory-mcp/README.md
@@ -32,13 +32,12 @@ npm install -g @usebruno/cli
 Since Memory MCP uses stdio, you'll need an HTTP bridge for Bruno testing:
 
 ```bash
-# Option 1: Use the MCP Inspector (recommended for debugging)
+# Use the MCP Inspector (recommended)
 npx @modelcontextprotocol/inspector python -m memory_mcp.server
-
-# Option 2: Custom HTTP bridge (if implemented)
-python -m memory_mcp.http_bridge --port 8080
 ```
 
+This opens an inspector UI at http://localhost:5173 with an MCP endpoint for testing.
+
 ### 3. Configure Environment
 
 The `environments/local.bru` file contains:
diff --git a/bruno/memory-mcp/Search Memory.bru b/bruno/memory-mcp/Search Memory.bru
@@ -52,7 +52,7 @@ docs {
   ## Search Types
 
   - **text**: Full-text search using SQLite FTS5
-  - **semantic**: Vector similarity search using FAISS
+  - **semantic**: Vector similarity search using embeddings
   - **hybrid**: Combines both methods for best results
 
   ## Response
diff --git a/config/mcp-servers.json b/config/mcp-servers.json
@@ -6,7 +6,7 @@
       "env": {
         "CLAUDE_CODE_PP_CONFIG": "~/.claude-code-pp/config/settings.yaml"
       },
-      "description": "Claude Code++ Memory System - tiered memory with Redis, FAISS, SQLite, and Obsidian vault"
+      "description": "Claude Code++ Memory System - tiered memory with Redis, Graphiti/Neo4j, SQLite, and Obsidian vault"
     },
     "claude-code-pp-system": {
       "command": "${HOME}/.claude-code-pp/bin/system-controller",
diff --git a/config/settings.yaml b/config/settings.yaml
@@ -56,20 +56,21 @@ memory:
   # Storage paths
   paths:
     vault: "~/.claude-code-pp/memory/vault"
-    faiss: "~/.claude-code-pp/memory/faiss"
     database: "~/.claude-code-pp/memory/metadata.db"
-  
+
   # Hot layer (Redis)
   redis:
     enabled: true
     host: "localhost"
     port: 6379
     db: 0
-  
-  # Warm layer (FAISS)
-  faiss:
-    index_type: "flat"  # flat | ivf | hnsw
-    rebuild_threshold: 0.1
+
+  # Warm layer (Graphiti/Neo4j)
+  graphiti:
+    enabled: true
+    neo4j_uri: "bolt://localhost:7687"
+    neo4j_user: "neo4j"
+    # neo4j_password: set via NEO4J_PASSWORD env var
   
   # Auto-memory triggers
   auto_store:
diff --git a/install.sh b/install.sh
@@ -202,7 +202,7 @@ create_directories() {
     info "Creating directory structure..."
 
     mkdir -p "$INSTALL_DIR"/{bin,config,logs,cache}
-    mkdir -p "$INSTALL_DIR"/memory/{sqlite,lancedb,vault/{code,notes,conversations,references,daily}}
+    mkdir -p "$INSTALL_DIR"/memory/{sqlite,vault/{code,notes,conversations,references,daily}}
 
     success "Created $INSTALL_DIR"
 }
diff --git a/python/memory_mcp/sqlite_index.py b/python/memory_mcp/sqlite_index.py
@@ -393,7 +393,7 @@ def get_recent(self, limit: int = 20) -> List[MemoryDocument]:
             return [MemoryDocument.from_row(row) for row in cursor.fetchall()]
 
     def get_by_embedding_ids(self, embedding_ids: List[str]) -> List[MemoryDocument]:
-        """Get documents by their embedding IDs (for FAISS result lookup)."""
+        """Get documents by their embedding IDs (for semantic search result lookup)."""
         if not embedding_ids:
             return []
         placeholders = ",".join("?" * len(embedding_ids))
diff --git a/python/pyproject.toml b/python/pyproject.toml
@@ -86,6 +86,7 @@ asyncio_mode = "auto"
 testpaths = ["tests"]
 markers = [
     "integration: marks tests as integration tests (require services)",
+    "slow: marks tests as slow (deselect with '-m \"not slow\"')",
 ]
 
 [tool.mypy]
diff --git a/python/requirements.txt b/python/requirements.txt
@@ -1,14 +1,10 @@
 # Memory MCP Server - Claude Code++
 # Core dependencies
 pyyaml>=6.0
-numpy>=1.24
 
 # Hot cache layer (optional but recommended)
 redis>=4.0
 
-# Vector search (optional but recommended)
-faiss-cpu>=1.7
-
 # Embedding providers (optional - needed for semantic search)
 httpx>=0.24
 
diff --git a/python/tests/test_backup_manager.py b/python/tests/test_backup_manager.py
@@ -146,7 +146,7 @@ def test_backup_creates_directory(self, local_backup_strategy, sample_metadata,
         local_backup_strategy.config.compress = False  # Disable compression
         sample_metadata.source_paths = {
             "sqlite": temp_source_dir["sqlite"],
-            "faiss": temp_source_dir["vault"]
+            "vault": temp_source_dir["vault"]
         }
 
         result = local_backup_strategy.backup(sample_metadata)
diff --git a/python/tests/test_cli_debug.py b/python/tests/test_cli_debug.py
@@ -279,11 +279,11 @@ def test_health_command_success(self, cli, capsys):
                         "sqlite": True,
                         "vault": True,
                         "redis": True,
-                        "faiss": True,
+                        "graphiti": True,
                         "embedder": True
                     },
                     "redis": {"cache_hits": 50, "cache_misses": 5},
-                    "faiss": {"total_vectors": 1000},
+                    "graphiti": {"entities": 500, "relationships": 1200},
                     "embedder": {"provider": "local/test"}
                 }),
                 "type": "text"
diff --git a/wiki/Configuration.md b/wiki/Configuration.md
@@ -96,7 +96,6 @@ MCP servers are configured in `~/.claude.json`:
 |----------|---------|-------------|
 | `SQLITE_PATH` | `~/.claude-code-pp/memory/sqlite/memories.db` | SQLite database path |
 | `OBSIDIAN_VAULT_PATH` | `~/.claude-code-pp/memory/vault` | Vault directory |
-| `LANCEDB_PATH` | `~/.claude-code-pp/memory/lancedb` | LanceDB directory |
 
 ### Redis
 
@@ -146,7 +145,6 @@ Default directory layout:
 ├── memory/
 │   ├── sqlite/
 │   │   └── memories.db     # SQLite database
-│   ├── lancedb/            # Vector embeddings
 │   └── vault/              # Obsidian-compatible notes
 │       ├── code/
 │       ├── notes/
diff --git a/wiki/Installation.md b/wiki/Installation.md
@@ -59,7 +59,7 @@ Install options:
 ### Step 3: Create Directory Structure
 
 ```bash
-mkdir -p ~/.claude-code-pp/{config,memory/{sqlite,lancedb,vault},logs,cache}
+mkdir -p ~/.claude-code-pp/{config,memory/{sqlite,vault},logs,cache}
 ```
 
 ### Step 4: Build Memory MCP Binary
@@ -131,8 +131,8 @@ docker-compose up -d
 ```
 
 This starts:
-- Redis (port 6379)
-- ChromaDB (port 8000) - Vector embeddings
+- Redis (port 6379) - Hot cache
+- Neo4j (port 7474/7687) - Knowledge graph
 - LiteLLM (port 4000) - Model routing
 
 With optional profiles:
diff --git a/wiki/Troubleshooting.md b/wiki/Troubleshooting.md
@@ -252,13 +252,7 @@ redis-cli CONFIG SET maxmemory 256mb
 redis-cli CONFIG SET maxmemory-policy allkeys-lru
 ```
 
-2. **Clear old embeddings:**
-```bash
-# Remove stale LanceDB data
-rm -rf ~/.claude-code-pp/memory/lancedb/*
-```
-
-3. **Vacuum SQLite:**
+2. **Vacuum SQLite:**
 ```bash
 sqlite3 ~/.claude-code-pp/memory/sqlite/memories.db "VACUUM;"
 ```
@@ -334,12 +328,6 @@ rm ~/.claude-code-pp/memory/sqlite/memories.db
 redis-cli FLUSHDB
 ```
 
-### Reset Embeddings
-
-```bash
-rm -rf ~/.claude-code-pp/memory/lancedb/*
-```
-
 ### Full Reset
 
 ```bash
@@ -348,7 +336,7 @@ cp -r ~/.claude-code-pp ~/.claude-code-pp.backup
 
 # Reset
 rm -rf ~/.claude-code-pp/memory/*
-mkdir -p ~/.claude-code-pp/memory/{sqlite,lancedb,vault}
+mkdir -p ~/.claude-code-pp/memory/{sqlite,vault}
 ```
 
 ## Getting Help

Original file line number	Diff line number	Diff line change
`@@ -202,7 +202,7 @@ create_directories() {`
`202`	`202`	`info "Creating directory structure..."`
`203`	`203`
`204`	`204`	`mkdir -p "$INSTALL_DIR"/{bin,config,logs,cache}`
`205`		`- mkdir -p "$INSTALL_DIR"/memory/{sqlite,lancedb,vault/{code,notes,conversations,references,daily}}`
	`205`	`+ mkdir -p "$INSTALL_DIR"/memory/{sqlite,vault/{code,notes,conversations,references,daily}}`
`206`	`206`
`207`	`207`	`success "Created $INSTALL_DIR"`
`208`	`208`	`}`