Skip to main content

Getting Started

Your second brain in 5 minutes. Connect any MCP-compatible AI assistant to ULPI Memory and start storing knowledge with cognitive decay and semantic search.
Prerequisites: Active ULPI account, repository created, MCP-compatible AI assistant (Claude Desktop, Cursor, Continue, Cline, Zed, Windsurf)

Quick Setup

1

Enable Memory Module

Dashboard → Repository → Settings → ModulesToggle “Enable Memory Module” → ONSystem initializes storage (~5 seconds) ✓
2

Choose Providers (Optional)

Embedding Provider:
  • Typesense (Recommended) - FREE, 384-dim, fast
  • OpenAI small - 1536-dim, costs tokens, better quality
  • OpenAI large - 3072-dim, costs more, best quality
Classification:
  • Regex (Recommended) - FREE, 85% accuracy
  • LLM (GPT-4) - Costs tokens, 95% accuracy
Cost-conscious? Use Typesense + Regex (both FREE). Switch later without re-embedding.
3

Get MCP Credentials

Admin Panel → MCP Setup → Copy Configuration
Save your API key securely. Shown only once. If lost, regenerate (invalidates old key).
4

Configure AI Assistant

Choose your platform below for setup instructions.
Config Location:
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • Linux: ~/.config/Claude/claude_desktop_config.json
Add MCP Server:
Restart: Quit (Cmd+Q / Alt+F4) and reopenVerify: “What MCP tools do you have?” → Lists 6 memory tools
5

Store First Memory

In your AI assistant:
Confirmation:
6

Search & Retrieve

Results:
Vector similarity understands “state management” relates to “useState” ✓

Verification

Ensure everything works:

MCP Connected

AI assistant shows “Connected to MCP servers”

Tools Available

“What MCP tools?” returns 6 memory tools:
  • store-memory
  • search-memories
  • retrieve-memory
  • reinforce-memory
  • prune-memories
  • delete-memory

Storage Works

Test memory stored with UUID confirmation

Search Works

Search returns relevant results

Troubleshooting

Symptoms: “MCP server offline” or no tools availableSolutions:
  1. Verify API key (no extra spaces)
  2. Check Tenant ID matches repo ID
  3. Test network: https://api.ulpi.io/health (200 OK)
  4. Validate JSON syntax (commas, brackets)
  5. Completely quit and restart AI assistant
  6. Check firewall allows HTTPS to api.ulpi.io
Check Logs: Look for MCP errors in assistant’s console
Symptoms: “Invalid API key” or “Authentication required”Solutions:
  1. Regenerate API key (Admin Panel → API Keys)
  2. Verify key format: starts with ulpi_
  3. Ensure header: Authorization: Bearer YOUR_KEY
  4. Check key scoped to correct tenant
  5. Verify not expired (1 year default)
Security: Never share API keys. If exposed, regenerate immediately.
Symptoms: store-memory succeeds but nothing in admin panelSolutions:
  1. Check correct repository in admin panel
  2. Verify response includes UUID
  3. Check embedding logs for errors
  4. Verify quota not exceeded (Starter: 1K, Pro: 10K)
  5. Check system status: https://status.ulpi.io
Debug: Test API directly: POST /api/v1/memories
Symptoms: Empty results despite memories existingSolutions:
  1. Wait 10-30 seconds for embedding generation
  2. Remove filters (sector, salience) to search all
  3. Use simple keywords first
  4. Check search index health (Admin → Stats)
  5. Lower min_salience (default: 0.3, try: 0.0)
Test: Search for exact content first, then semantic
Symptoms: “I don’t have access to memory tools”Solutions:
  1. Completely quit and reopen assistant
  2. Verify correct config file path
  3. Validate JSON syntax (jsonlint.com)
  4. Check assistant version supports MCP
  5. Verify file permissions (readable)
  6. Test server: curl https://api.ulpi.io/mcp/memory
Check Logs: Debug logs for MCP connection errors
Symptoms: Operations take 10+ seconds or timeoutSolutions:
  1. Check internet speed (run speed test)
  2. Disable waypoint expansion (expand_waypoints: false)
  3. Reduce search limit (limit: 5)
  4. Switch to Typesense (faster than OpenAI)
  5. Check https://status.ulpi.io
Optimize: Use filters (sector, date range) before ranking

Explore Features

Reinforce Memories

Δ0.15 salience boost, slows decay

Filter by Sector

Returns only Semantic (facts) memories

Retrieve by ID

Direct access to known memories

Prune Old Memories

Remove faded memories

Quick Reference

6 MCP Tools

Common Commands


Next Steps

How It Works

Cognitive science behind the system

Cognitive Sectors

5 memory types and when to use each

Semantic Waypoints

Automatic context expansion

Search & Retrieval

Master hybrid search techniques

Decay & Reinforcement

Control memory lifespan

Best Practices

Optimization tips and strategies

Support

Documentation

Complete guides and references

Admin Panel

Manage memories and settings

Email Support

support@ulpi.io
  • Starter: 48 hours
  • Pro: 24 hours priority
  • Enterprise: 4 hours dedicated

Community

  • Forum: forum.ulpi.io
  • GitHub: github.com/ulpi-io/ulpi/issues

Your second brain is ready. Start storing memories and watch semantic connections emerge naturally.