memory-pulse-mcp-server 0.1.15 → 0.1.17

package/README.md

@@ -14,6 +14,33 @@
 
 ---
 
+## What's New
+
+### v0.1.16 - Smart Retrieval with Auto-Degradation
+
+This release introduces a major upgrade to the retrieval engine. **Memory Pulse now guarantees you always get results** -- even when your query keywords don't exactly match stored content.
+
+**Multi-Dimension Search** -- Searches across 5 dimensions simultaneously: `projectId`, `summary`, `fullText`, `tags`, and `keywords`. Any keyword matching any dimension counts as a hit.
+
+**Smart Auto-Degradation** -- When no results are found, the engine automatically relaxes search conditions in two stages:
+
+```
+Stage 1: Query matches nothing?
+         -> Remove keywords, return latest memories for the project
+
+Stage 2: Still nothing?
+         -> Remove project filter, search globally
+```
+
+This means `mpulse_recall` with a `projectId` will **never return empty** as long as that project has any memories stored -- no matter what query you use.
+
+**Other improvements:**
+- Auto-relationship building: new memories automatically link to related existing memories
+- Improved PostgreSQL storage: `multiDimensionSearch` replaces legacy search methods
+- Fixed `storeSolution` tags handling for PostgreSQL
+
+---
+
 ## Why Memory Pulse?
 
 ### The Problem with Existing AI Memory Solutions
@@ -30,26 +57,31 @@
 | Feature | Memory Pulse | mem0 | Zep | LangChain Memory |
 |---------|-------------|------|-----|------------------|
 | **Storage** | Full context (no compression) | Vector embeddings | Vector + Graph | Vector embeddings |
-| **Retrieval Strategy** | L1 Exact → L2 Full-text → L3 Semantic | Semantic only | Semantic + Temporal | Semantic only |
+| **Retrieval** | Multi-dimension + auto-degradation | Semantic only | Semantic + Temporal | Semantic only |
+| **Never Empty Recall** | Yes (auto-degradation) | No | No | No |
 | **Decision Tracking** | Forced structured fields | - | - | - |
+| **Auto-Relationships** | Built-in | - | - | - |
 | **Relationship Graph** | Built-in | - | Built-in | - |
 | **Timeline View** | Built-in | - | Built-in | - |
 | **MCP Native** | Yes | - | - | - |
-| **Local-first** | SQLite | Cloud-dependent | Cloud-dependent | Varies |
+| **Local-first** | SQLite + PostgreSQL | Cloud-dependent | Cloud-dependent | Varies |
 | **Zero Config** | Yes | - | - | - |
 
 ---
 
 ## Features
 
+- **Always Gets Results** - Smart auto-degradation ensures you never get empty recalls
+- **Multi-Dimension Search** - Searches across projectId, summary, fullText, tags, and keywords simultaneously
 - **Complete Context Preservation** - No compression, no information loss
-- **Multi-level Retrieval** - L1 Exact match → L2 Full-text search → L3 Semantic search
 - **Structured Memory Types** - Decision / Solution / Session / Code / Error / Config
+- **Auto-Relationship Building** - New memories automatically link to related existing ones
 - **Relationship Chains** - Track how memories relate and evolve
 - **Timeline View** - See memory evolution over time
 - **Forced Structure** - AI must provide complete context (no lazy summaries)
-- **Local-first** - SQLite storage, your data stays with you
-- **Zero Config** - Works out of the box
+- **Local-first** - SQLite storage by default, your data stays with you
+- **PostgreSQL Support** - Scale to cloud for team usage
+- **Zero Config** - Works out of the box with `npx`
 
 ---
 
@@ -337,38 +369,45 @@ Relationships:
 
 ## Retrieval Algorithm
 
-Memory Pulse uses a **3-level cascade retrieval strategy**:
+Memory Pulse uses a **multi-dimension search with smart auto-degradation**:
 
 ```
-┌─────────────────────────────────────────────────────┐
-│                    User Query                        │
-└─────────────────────┬───────────────────────────────┘
-                      ▼
-┌─────────────────────────────────────────────────────┐
-│  L1: Exact Match (< 10ms)                           │
-│  - projectId + type + keywords index                │
-│  - Returns if matches ≥ 5                           │
-└─────────────────────┬───────────────────────────────┘
-                      ▼ (if insufficient)
-┌─────────────────────────────────────────────────────┐
-│  L2: Full-text Search (< 100ms)                     │
-│  - SQLite FTS5 / PostgreSQL full-text               │
-│  - Chinese + English tokenization                   │
-│  - Returns if matches ≥ 3                           │
-└─────────────────────┬───────────────────────────────┘
-                      ▼ (if insufficient)
-┌─────────────────────────────────────────────────────┐
-│  L3: Semantic Search (< 500ms)                      │
-│  - Embedding similarity (optional)                  │
-│  - Fallback for fuzzy queries                       │
-└─────────────────────────────────────────────────────┘
+┌──────────────────────────────────────────────────────┐
+│                     User Query                        │
+│            (query + projectId + filters)              │
+└──────────────────────┬───────────────────────────────┘
+                       ▼
+┌──────────────────────────────────────────────────────┐
+│  Multi-Dimension Search (< 100ms)                     │
+│  Searches 5 dimensions simultaneously:               │
+│  projectId | summary | fullText | tags | keywords    │
+│  Any keyword matching any dimension = hit            │
+└──────────────────────┬───────────────────────────────┘
+                       ▼ (if 0 results)
+┌──────────────────────────────────────────────────────┐
+│  Auto-Degradation Stage 1                             │
+│  Remove query keywords, keep projectId               │
+│  -> Returns latest memories for the project          │
+└──────────────────────┬───────────────────────────────┘
+                       ▼ (if still 0 results)
+┌──────────────────────────────────────────────────────┐
+│  Auto-Degradation Stage 2                             │
+│  Remove projectId filter, search globally            │
+│  -> Returns latest memories across all projects      │
+└──────────────────────┬───────────────────────────────┘
+                       ▼ (optional, if vector available)
+┌──────────────────────────────────────────────────────┐
+│  Semantic Search Fallback                             │
+│  Embedding similarity for fuzzy matching             │
+└──────────────────────────────────────────────────────┘
 ```
 
 **Why this approach?**
 
-- **Precision first**: Exact matches are faster and more accurate
-- **Graceful degradation**: Falls back to broader search when needed
-- **No false positives**: Semantic search is last resort, not default
+- **Never empty**: Smart degradation guarantees results when memories exist
+- **Precision first**: Multi-dimension exact matching before relaxing conditions
+- **Graceful fallback**: Progressively broader search, not a sudden jump to fuzzy matching
+- **Works with any language**: No tokenizer dependency, LIKE-based matching works with Chinese, English, and mixed content
 
 ---
 
@@ -415,8 +454,9 @@ Features:
 - [x] MCP Server core functionality
 - [x] SQLite local storage
 - [x] PostgreSQL cloud support
-- [x] Multi-level retrieval (Exact + Full-text)
+- [x] Multi-dimension retrieval with smart auto-degradation
 - [x] Decision/Solution/Session structured storage
+- [x] Auto-relationship building between memories
 - [x] Web Dashboard for visualization
 - [ ] CLI tool for manual operations
 - [ ] Semantic search (Embedding)