feat: Add session memory system with pgvector

- Add session_memories table for persistent learnings
- Add memory_add, memory_search, memory_list, memory_context tools
- Supports semantic search via embeddings
- Categories: pattern, fix, preference, gotcha, architecture

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

README.md

@@ -4,12 +4,13 @@ Exposes task management tools via Model Context Protocol. Uses PostgreSQL with p
 
 ## Requirements
 
-- SSH tunnel to docker-host: `ssh -L 5435:localhost:5435 docker-host -N &`
-- PostgreSQL with pgvector on docker-host (litellm-pgvector container)
+- SSH tunnel to docker-host: `ssh -L 5432:localhost:5432 docker-host -N &`
+- PostgreSQL with pgvector on docker-host (CI stack `postgres` container)
+- CI postgres must be running: `ssh docker-host "cd /opt/docker/ci && docker compose up -d postgres"`
 
 ## Configuration
 
-Add to `~/.claude/mcp_servers.json`:
+Add to `~/.claude/settings.json` under `mcpServers`:
 
 ```json
 {
@@ -18,10 +19,10 @@ Add to `~/.claude/mcp_servers.json`:
     "args": ["/path/to/task-mcp/dist/index.js"],
     "env": {
       "DB_HOST": "localhost",
-      "DB_PORT": "5435",
-      "DB_NAME": "litellm",
-      "DB_USER": "litellm",
-      "DB_PASSWORD": "litellm",
+      "DB_PORT": "5432",
+      "DB_NAME": "agiliton",
+      "DB_USER": "agiliton",
+      "DB_PASSWORD": "<from /opt/docker/ci/.env>",
       "LLM_API_URL": "https://llm.agiliton.cloud",
       "LLM_API_KEY": "sk-master-..."
     }