fix: resolve remaining merge conflicts in graph.js and store interface

Author: Skiipy11

.env.example

@@ -28,10 +28,16 @@ QDRANT_API_KEY=your-qdrant-key-here
 EMBEDDING_PROVIDER=openai
 OPENAI_API_KEY=your-openai-key-here
 # For Gemini:
+<<<<<<< HEAD
 # EMBEDDING_PROVIDER=gemini
 # GEMINI_API_KEY=your-gemini-key-here
 # GEMINI_EMBEDDING_MODEL=gemini-embedding-2-preview  # default
 # GEMINI_EMBEDDING_DIMS=3072                          # default (also supports 1536, 768 via Matryoshka)
+=======
+# GEMINI_API_KEY=your-gemini-key-here
+# GEMINI_EMBEDDING_MODEL=gemini-embedding-exp-03-07
+# GEMINI_EMBEDDING_DIMS=3072  # 3072 (default), 1536, or 768
+>>>>>>> worktree-agent-a1f30793
 # For Ollama:
 # OLLAMA_URL=http://localhost:11434
 # OLLAMA_MODEL=nomic-embed-text
@@ -66,6 +72,14 @@ DECAY_FACTOR=0.98
 # --- Event TTL ---
 # EVENT_TTL_DAYS=30  # Auto-expire low-importance, never-accessed events after N days (default: 30)
 
+# --- Client Resolver ---
+# BASEROW_CLIENTS_TABLE_ID=734  # Baserow table with client_fingerprints field
+# BASEROW_CLIENT_TOKEN=         # Uses BASEROW_API_KEY if not set
+
+# --- Webhook Notifications ---
+# WEBHOOK_URLS=https://n8n.example.com/webhook/brain  # Comma-separated webhook URLs for memory events
+# WEBHOOK_EVENTS=store,supersede,delete               # Which events trigger notifications (default: all)
+
 # --- MCP Server Timeouts ---
 # BRAIN_MCP_TIMEOUT=15000          # Default timeout for MCP→API calls in ms (default: 15000)
 # BRAIN_MCP_CONSOLIDATION_TIMEOUT=120000  # Timeout for sync consolidation calls in ms (default: 120000)

README.md

@@ -187,11 +187,19 @@ This means you get both "find memories similar to X" *and* "give me all facts wi
 │                        Memory API (Express)                               │
 │  POST /memory  GET /memory/search  GET /briefing  GET /stats              │
 │  GET /memory/query  POST /webhook/n8n  POST /consolidate  GET /entities   │
+│  GET /client/:name  GET /export  POST /import  GET /graph                 │
 ├──────────────────────┬────────────────────────────────────────────────────┤
 │   Embedding Layer    │            LLM Layer                               │
+<<<<<<< HEAD
 │  ┌────────┐ ┌──────┐ ┌──────┐│  ┌────────┐ ┌───────────┐ ┌──────┐ ┌──────┐│
 │  │ OpenAI │ │Gemini│ │Ollama││  │ OpenAI │ │ Anthropic │ │Gemini│ │Ollama││
 │  └────────┘ └──────┘ └──────┘│  └────────┘ └───────────┘ └──────┘ └──────┘│
+=======
+│  ┌────────┐ ┌──────┐│  ┌────────┐ ┌───────────┐ ┌──────┐ ┌──────┐      │
+│  │ OpenAI │ │Gemini││  │ OpenAI │ │ Anthropic │ │Gemini│ │Ollama│      │
+│  │        │ │Ollama││  └────────┘ └───────────┘ └──────┘ └──────┘      │
+│  └────────┘ └──────┘│                                                    │
+>>>>>>> worktree-agent-a1f30793
 ├──────────────────────┴────────────────────────────────────────────────────┤
 │                          Storage Layer                                    │
 │  ┌────────────────────┐  ┌────────┐ ┌────────┐ ┌───────┐                │
@@ -245,6 +253,7 @@ Entities are automatically extracted from the content and stored in both the Qdr
 | `category` | No | `semantic`, `episodic`, or `procedural`. Default: `episodic` |
 | `importance` | No | `critical`, `high`, `medium`, or `low`. Default: `medium` |
 | `key` | No | For facts: unique key enabling upsert |
+| `knowledge_category` | No | `brand`, `strategy`, `meeting`, `content`, `technical`, `relationship`, or `general`. Auto-classified during consolidation if not set. |
 | `subject` | No | For statuses: what system this status is about |
 | `status_value` | No | For statuses: the current status string |
 
@@ -412,6 +421,63 @@ Entity types: `client`, `person`, `system`, `service`, `domain`, `technology`, `
 
 Requires SQLite or Postgres structured storage backend. Baserow users get entity data in Qdrant payloads but not the entity graph endpoints.
 
+### `GET /client/:name` — Client briefing
+
+```bash
+curl "http://localhost:8084/client/acme-corp" -H "X-Api-Key: YOUR_KEY"
+```
+
+Returns a comprehensive client briefing: all memories tagged with the resolved client_id, grouped by knowledge_category (brand, strategy, meeting, content, technical, relationship, general). Supports fuzzy name resolution — "acme", "Acme Corp", and "acme-corp" all resolve to the same client.
+
+| Param | Description |
+|-------|-------------|
+| `name` | Client name or slug (fuzzy matched against fingerprints) |
+| `category` | Filter by knowledge_category |
+| `limit` | Max memories per category (default 20) |
+
+### `GET /export` — Export memories
+
+```bash
+curl "http://localhost:8084/export?format=json" -H "X-Api-Key: YOUR_KEY" -o backup.json
+```
+
+Exports all active memories as JSON. Useful for backup before embedding provider migration or system upgrades.
+
+| Param | Description |
+|-------|-------------|
+| `format` | `json` (default) |
+| `client_id` | Filter export to a specific client |
+| `type` | Filter by memory type |
+
+### `POST /import` — Import memories
+
+```bash
+curl -X POST "http://localhost:8084/import" \
+  -H "Content-Type: application/json" \
+  -H "X-Api-Key: YOUR_KEY" \
+  -d @backup.json
+```
+
+Imports memories from a previous export. Handles deduplication (skips exact hash matches) and batch processes embeddings. Safe for embedding provider migration — re-embeds all content with the current provider.
+
+### `GET /graph` — Entity relationship graph
+
+```bash
+# JSON data
+curl "http://localhost:8084/graph?format=json" -H "X-Api-Key: YOUR_KEY"
+
+# Interactive D3.js visualization (open in browser)
+# http://localhost:8084/graph
+```
+
+Returns entity relationships as a node/edge graph. Co-occurrence relationships are automatically detected during consolidation. The HTML format serves an interactive D3.js force-directed visualization with dark theme, search, and zoom.
+
+| Param | Description |
+|-------|-------------|
+| `format` | `html` (default, interactive viz) or `json` (raw data) |
+| `entity` | Center graph on a specific entity |
+| `depth` | Relationship traversal depth (default 2) |
+
 ### Backfill Existing Memories
 
 After upgrading to v1.2.0, run the backfill script once to extract entities from existing memories:
@@ -445,7 +511,7 @@ docker exec shared-brain-api node scripts/reindex-embeddings.js
 
 ### MCP Server (Claude Code, Cursor, Windsurf)
 
-The MCP server exposes 8 tools: `brain_store`, `brain_search`, `brain_briefing`, `brain_query`, `brain_stats`, `brain_consolidate`, `brain_entities`, `brain_delete`.
+The MCP server exposes 12 tools: `brain_store`, `brain_search`, `brain_briefing`, `brain_query`, `brain_stats`, `brain_consolidate`, `brain_entities`, `brain_delete`, `brain_client`, `brain_export`, `brain_import`, `brain_graph`.
 
 **Claude Code (`~/.claude.json`):**
 ```json
@@ -553,8 +619,13 @@ All configuration is via environment variables. Copy `.env.example` to `.env` an
 | `EMBEDDING_PROVIDER` | `openai` | `openai`, `gemini`, or `ollama` |
 | `OPENAI_API_KEY` | — | Required when using OpenAI embeddings |
 | `GEMINI_API_KEY` | — | Required when using Gemini embeddings |
+<<<<<<< HEAD
 | `GEMINI_EMBEDDING_MODEL` | `gemini-embedding-2-preview` | Gemini embedding model name |
 | `GEMINI_EMBEDDING_DIMS` | `3072` | Output dimensions (3072, 1536, or 768 via Matryoshka) |
+=======
+| `GEMINI_EMBEDDING_MODEL` | `gemini-embedding-exp-03-07` | Gemini embedding model |
+| `GEMINI_EMBEDDING_DIMS` | `3072` | Output dimensions (3072, 1536, or 768 — Matryoshka support) |
+>>>>>>> worktree-agent-a1f30793
 | `OLLAMA_URL` | `http://localhost:11434` | Ollama server URL |
 | `OLLAMA_MODEL` | `nomic-embed-text` | Ollama embedding model name |
 
@@ -590,6 +661,24 @@ Entity graph tables (entities, aliases, memory links) are automatically created
 
 Only affects `fact` and `status` types. Events and decisions don't decay.
 
+### Client Resolver
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `BASEROW_CLIENTS_TABLE_ID` | — | Baserow table ID containing client records with `client_fingerprints` field |
+| `BASEROW_CLIENT_TOKEN` | — | API token for client table access. Falls back to `BASEROW_API_KEY` if not set |
+
+When configured, the client resolver auto-tags `client_id` on incoming memories by matching content against client fingerprints (accent-normalized, fuzzy). This enables automatic client attribution without agents needing to specify `client_id` explicitly.
+
+### Webhook Notifications
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `WEBHOOK_URLS` | — | Comma-separated webhook URLs for memory event notifications |
+| `WEBHOOK_EVENTS` | `store,supersede,delete` | Which events trigger notifications |
+
+When configured, the system dispatches real-time HTTP POST notifications to the specified URLs on memory store, supersede, and delete events. Useful for triggering n8n workflows, Slack alerts, or external dashboards.
+
 ## Deployment
 
 ### Docker (Recommended)
@@ -644,12 +733,19 @@ multi-agent-memory/
 │   │   │   ├── stats.js        # Memory + entity health dashboard
 │   │   │   ├── consolidation.js# Consolidation trigger/status
 │   │   │   ├── webhook.js      # n8n webhook with entity extraction
-│   │   │   └── entities.js     # Entity graph endpoints
+│   │   │   ├── entities.js     # Entity graph endpoints
+│   │   │   ├── client.js       # Client briefing endpoint
+│   │   │   ├── export.js       # Import/export endpoints
+│   │   │   └── graph.js        # Entity relationship graph + D3.js viz
+│   │   ├── templates/
+│   │   │   └── graph.html      # Interactive D3.js force-directed graph
 │   │   └── services/
 │   │       ├── qdrant.js       # Vector store + entity index
 │   │       ├── scrub.js        # Credential scrubbing
 │   │       ├── entities.js     # Entity extraction, alias cache, linking
 │   │       ├── consolidation.js# LLM consolidation + entity refinement
+│   │       ├── client-resolver.js # Fingerprint-based client identification
+│   │       ├── notifications.js# Webhook notification dispatch
 │   │       ├── embedders/      # Pluggable embedding providers
 │   │       │   ├── interface.js
 │   │       │   ├── openai.js
@@ -673,7 +769,7 @@ multi-agent-memory/
 │   ├── Dockerfile
 │   └── package.json
 ├── mcp-server/                 # MCP server for Claude/Cursor
-│   ├── src/index.js            # 8 tools: store, search, briefing, query, stats, consolidate, entities, delete
+│   ├── src/index.js            # 12 tools: store, search, briefing, query, stats, consolidate, entities, delete, client, export, import, graph
 │   └── package.json
 ├── adapters/
 │   ├── bash/                   # CLI adapter (curl + jq)
@@ -689,12 +785,10 @@ multi-agent-memory/
 ## Roadmap
 
 - **Web dashboard** — Browse, search, and manage memories visually
-- **Entity relationships** — Typed edges between entities (e.g., "Client X uses Technology Y")
 - **Python SDK** — `pip install multi-agent-memory`
 - **Automatic memory capture** — System learns what's worth remembering vs what's noise
 - **Multi-collection support** — Isolated memory spaces per project or team
-- **Real-time notifications** — SSE/WebSocket for agents to subscribe to memory updates
-- **Memory import/export** — Bulk operations for backup and migration
+- **SSE/WebSocket subscriptions** — Real-time streaming for agents to subscribe to memory updates
 
 ## Contributing
 

api/src/routes/graph.js

@@ -1,59 +1,4 @@
 import { Router } from 'express';
-<<<<<<< HEAD
-import { findEntity, getRelationships } from '../services/stores/interface.js';
-import { isEntityStoreAvailable } from '../services/stores/interface.js';
-
-export const graphRouter = Router();
-
-/**
- * GET /graph/:entity — Query entity relationships
- * Query params:
- *   - min_strength (default 1): minimum relationship strength
- *   - depth (default 1, max 3): how many hops to traverse
- */
-graphRouter.get('/:entity', async (req, res) => {
-  try {
-    if (!isEntityStoreAvailable()) {
-      return res.status(501).json({
-        error: 'Entity graph requires sqlite or postgres backend. Set STRUCTURED_STORE in .env.',
-      });
-    }
-
-    const entityName = req.params.entity;
-    const minStrength = parseInt(req.query.min_strength) || 1;
-    const depth = Math.min(Math.max(parseInt(req.query.depth) || 1, 1), 3);
-
-    const entity = await findEntity(entityName);
-    if (!entity) {
-      return res.status(404).json({ error: `Entity not found: ${entityName}` });
-    }
-
-    const rootEntity = {
-      id: entity.id,
-      name: entity.canonical_name,
-      type: entity.entity_type,
-    };
-
-    const relationships = await getRelationships(entity.id, minStrength);
-
-    if (depth === 1) {
-      return res.json({
-        entity: rootEntity,
-        relationships,
-        depth,
-      });
-    }
-
-    // For depth > 1, recursively fetch relationships of connected entities
-    const visited = new Set([entity.id]);
-    const enriched = await enrichRelationships(relationships, minStrength, depth - 1, visited);
-
-    return res.json({
-      entity: rootEntity,
-      relationships: enriched,
-      depth,
-    });
-=======
 import { readFileSync } from 'fs';
 import { fileURLToPath } from 'url';
 import { dirname, join } from 'path';
@@ -123,73 +68,12 @@ graphRouter.get('/:entity', async (req, res) => {
 
     const graphData = await buildGraphData(entity.canonical_name, depth);
     res.json(graphData);
->>>>>>> worktree-agent-a801b191
   } catch (err) {
     console.error('[graph] Error:', err.message);
     res.status(500).json({ error: err.message });
   }
 });
 
-<<<<<<< HEAD
-/**
- * Recursively fetch nested relationships up to remaining depth.
- * Uses a visited set to avoid cycles.
- */
-async function enrichRelationships(relationships, minStrength, remainingDepth, visited) {
-  if (remainingDepth <= 0) return relationships;
-
-  const enriched = [];
-  for (const rel of relationships) {
-    const enrichedRel = { ...rel };
-    if (!visited.has(rel.entity.id)) {
-      visited.add(rel.entity.id);
-      const nested = await getRelationships(rel.entity.id, minStrength);
-      // Filter out already-visited entities from nested results
-      const filtered = nested.filter(n => !visited.has(n.entity.id));
-      enrichedRel.relationships = await enrichRelationships(filtered, minStrength, remainingDepth - 1, visited);
-    } else {
-      enrichedRel.relationships = [];
-    }
-    enriched.push(enrichedRel);
-  }
-  return enriched;
-}
-
-/**
- * GET /graph/:entity/html — Placeholder for D3.js visualization (Task 11)
- */
-graphRouter.get('/:entity/html', async (req, res) => {
-  if (!isEntityStoreAvailable()) {
-    return res.status(501).json({
-      error: 'Entity graph requires sqlite or postgres backend. Set STRUCTURED_STORE in .env.',
-    });
-  }
-
-  const entityName = req.params.entity;
-  res.setHeader('Content-Type', 'text/html');
-  res.send(`<!DOCTYPE html>
-<html lang="en">
-<head>
-  <meta charset="UTF-8">
-  <meta name="viewport" content="width=device-width, initial-scale=1.0">
-  <title>Graph: ${entityName}</title>
-  <style>
-    body { font-family: system-ui, sans-serif; display: flex; align-items: center; justify-content: center; height: 100vh; margin: 0; background: #1a1a2e; color: #e0e0e0; }
-    .loading { text-align: center; }
-    .loading h1 { font-size: 1.5rem; margin-bottom: 0.5rem; }
-    .loading p { color: #888; }
-  </style>
-</head>
-<body>
-  <div class="loading">
-    <h1>Graph visualization loading...</h1>
-    <p>Entity: <strong>${entityName}</strong></p>
-    <p>Full D3.js interactive visualization coming in the next update.</p>
-  </div>
-</body>
-</html>`);
-});
-=======
 // --- Graph building ---
 
 /**
@@ -346,4 +230,3 @@ function escapeHtml(str) {
     .replace(/"/g, '&quot;')
     .replace(/'/g, '&#039;');
 }
->>>>>>> worktree-agent-a801b191

api/src/services/stores/interface.js

@@ -113,7 +113,6 @@ export async function getEntityStats() {
   return requireStore().getEntityStats();
 }
 
-<<<<<<< HEAD
 export async function createRelationship(sourceId, targetId, type) {
   return requireStore().createRelationship(sourceId, targetId, type);
 }
@@ -124,9 +123,9 @@ export async function getRelationships(entityId, minStrength) {
 
 export async function listRelationships(filters) {
   return requireStore().listRelationships(filters);
-=======
+}
+
 // Direct store access (used by graph visualization for co-occurrence queries)
 export function _getStoreInstance() {
   return store;
->>>>>>> worktree-agent-a801b191
 }

mcp-server/CHANGELOG.md

@@ -1,5 +1,23 @@
 # Changelog
 
+## 2.0.0 (2026-03-20)
+
+### Features
+- **Client knowledge base**: Fingerprint-based client identification with accent normalization, `knowledge_category` field (brand/strategy/meeting/content/technical/relationship/general), `brain_client` tool for one-call client briefings with fuzzy name resolution
+- **Import/Export**: `brain_export` and `brain_import` tools for backup and embedding migration safety, with dedup and batch processing
+- **Webhook notifications**: Real-time dispatch on memory store/supersede/delete events via configurable webhook URLs
+- **Entity graph**: Relationship tracking with co-occurrence detection, `brain_graph` tool, interactive D3.js visualization (dark theme, force-directed, searchable)
+- **Consolidation enhancements**: Automatic knowledge_category reclassification and entity relationship type classification during 6h consolidation pass
+- **Auto-resolve client_id**: Memory store auto-tags client_id from content using fingerprint matching when not explicitly provided
+- **Gemini Embedding 2**: New pluggable embedder with task-type-aware embeddings (RETRIEVAL_DOCUMENT/RETRIEVAL_QUERY), Matryoshka support (3072/1536/768 dims)
+
+## 1.5.0
+
+### Long-Term Memory Hygiene
+- **Access-weighted search** — search results factor in access count alongside similarity and confidence, rewarding frequently-accessed memories
+- **Insight removal** — consolidation-generated insights can now be removed when source memories are deleted
+- **Entity fix** — fixed entity extraction for memories with no client_id
+
 ## 1.4.0
 
 ### Token Optimization