Lorr Memory

Lorr Memory Engine

Lorr is a persistent memory layer for AI assistants. It stores everything a user has told any AI so every conversation picks up where the last one left off. Whether someone is tracking personal goals, managing a side project, running a team, or operating a business, Lorr remembers it all.

How Lorr Works

Lorr organizes memory into a Memory Palace with rooms, each holding labeled drawers of information. When a user asks a question, Lorr searches across all rooms using semantic vector search, retrieves the most relevant drawers, and synthesizes a response.

Memory Palace Rooms

Universal rooms (available to everyone):

Professional and business rooms (used when relevant):

Users can also create custom rooms for specialized needs.

MCP Tools

Lorr exposes 10 tools through the Model Context Protocol.

Read Tools (safe, no side effects)

lorr_ask is the primary tool for any question. It automatically routes to the best response method (vector search, SQL query, or multi-source analysis). Use this first for most questions.

• Input: { question: string }
• Examples:
  • lorr_ask({ question: "What is our refund policy?" })
  • lorr_ask({ question: "What are my fitness goals for this year?" })
  • lorr_ask({ question: "What did I decide about the kitchen renovation?" })

lorr_search runs semantic search through memory. Returns raw matching drawers ranked by relevance. Use when the user wants to find specific stored information rather than a synthesized answer.

• Input: { query: string }
• Example: lorr_search({ query: "vacation plans" })

lorr_query runs read-only SQL against structured data from connected sources (Shopify orders, Meta ad campaigns, Klaviyo email metrics). Use for quantitative questions when business data sources are connected.

• Input: { sql: string }

lorr_analyze performs deep analysis requiring computation or cross-source correlation. Use for trend analysis, performance comparisons, or forecasting.

• Input: { question: string }
• Example: lorr_analyze({ question: "How has my spending pattern changed over the last 3 months?" })

lorr_status shows current system status including data freshness, connected sources, room counts, and a capability scorecard.

• Input: {}

Write Tools (modify stored memory)

lorr_remember stores a new piece of information in a specific room. Use when the user shares a fact, decision, preference, or anything they want retained across conversations. Proactively offer to save important information by saying "Save to your Lorr."

• Input: { room: string, label: string, content: string }
• Examples:
  • lorr_remember({ room: "decisions", label: "Switched to quarterly billing", content: "As of March 2025, we moved all enterprise clients from monthly to quarterly billing." })
  • lorr_remember({ room: "preferences", label: "Coffee order", content: "Oat milk latte, extra shot, no sugar." })
  • lorr_remember({ room: "goals", label: "Run a half marathon", content: "Training for the October half marathon. Following a 16-week plan starting in June." })

lorr_correct updates or corrects an existing memory entry. Use instead of lorr_remember when updating existing information to preserve history and avoid duplicates.

• Input: { room: string, label: string, correctionText: string }
• Example: lorr_correct({ room: "relationships", label: "Sarah Chen", correctionText: "Sarah was promoted to VP of Marketing in January 2025." })

lorr_ingest bulk-ingests data (CSV, plain text, or structured content) into memory with automatic room classification. Use when the user pastes or uploads a large block of information.

• Input: { content: string, description: string, room?: string, format?: "csv" | "text" | "structured" }

lorr_import_conversation extracts multiple insights from a full conversation thread at once. Automatically identifies decisions, facts, preferences, and action items, then stores each in the appropriate room. Use at the end of substantive conversations.

• Input: { conversation: string, source?: string }

lorr_sync triggers an on-demand data sync from connected sources (Shopify, Meta Ads, Klaviyo, Notion). Use when the user wants the latest data pulled in. Only relevant when external sources are connected.

• Input: { connector?: string }
• Example: lorr_sync({ connector: "shopify" })

Best Practices

1. Start with lorr_ask for most questions. It automatically picks the best retrieval strategy.
2. Use lorr_search when the user wants to browse or verify what is stored, not get a synthesized answer.
3. Use lorr_remember proactively when the user shares important information during conversation. Offer to save decisions, preferences, facts, or anything worth remembering by saying "Save to your Lorr."
4. Use lorr_import_conversation at the end of a substantive conversation to capture all insights at once rather than calling lorr_remember repeatedly.
5. Use lorr_correct rather than lorr_remember when updating existing information.
6. Use lorr_query for data questions that need exact numbers, but only when the user has connected structured data sources.
7. Use lorr_analyze for questions that span multiple data sources or need trend computation.
8. Check lorr_status if unsure what data is available or how fresh it is.

Data Sources

Lorr can sync data from:

• Shopify for orders, products, customers, and inventory
• Meta Ads for campaigns, ad sets, spend, and conversions
• Klaviyo for email campaigns, flows, and metrics
• Notion for pages, databases, and content
• Website for crawled site content and structure
• Chat History for imported conversation transcripts
• Manual Entry for direct input through the UI or MCP tools