Limitless CLI Skill

A comprehensive CLI tool for interacting with Limitless.ai Pendant data, featuring FalkorDBLite semantic graph with vector embeddings, content extraction, and DAG pipelines.

When to Use

• User wants semantic search across lifelogs, chats, persons, or topics
• User wants to query lifelogs or personal memory data
• User wants to extract entities (speakers, topics) from conversations
• User wants to sync lifelogs to a graph database
• User wants to run extraction or analysis pipelines
• User mentions "Limitless", "pendant", "lifelog", or "personal memory"
• User needs domain-specific extraction (medical, technical, business)

Quick Start

Semantic Search (Recommended)

# Semantic search across lifelogs (vector-based similarity)
limitless semantic-search "ICU critical care discussion" --types Lifelog --scores

# Search multiple node types
limitless semantic-search "family doctor" --types Lifelog,Chat,Person --limit 10

# Hybrid search (semantic + full-text)
limitless search "medical exam" --mode hybrid

# Check index status (embeddings, node counts)
limitless index status

Graph Queries

# Cypher query on semantic graph
limitless graph query "MATCH (p:Person)-[:SPOKE_IN]->(l:Lifelog) RETURN p.name, count(l) ORDER BY count(l) DESC LIMIT 10"

# Graph statistics
limitless graph stats

Basic Operations

# List recent lifelogs
limitless lifelogs list --limit 10

# Search for topic (full-text)
limitless lifelogs search "topic"

# Get specific lifelog
limitless lifelogs get <id> --format json

Architecture

FalkorDBLite Semantic Graph

The CLI uses an embedded FalkorDBLite graph database (no Docker required) with vector embeddings for semantic search.

Graph Statistics (as of 2026-01-11):

• 95,021 nodes (8,777 Lifelogs, 2,608 Chats, 1,549 Persons, 33,024 Topics, 47,027 Speakers)
• 81,356 relationships (SPOKE_IN, HAS_TOPIC, HAS_CONTACT)
• 100% embedding coverage on searchable nodes

Core Commands

References (Load on Demand)

• API Reference: Rate limiting (180 req/min), retry logic
• Database Schema: Node types, relationships, vector indexes
• Extraction Rules: Rule-based and LLM extraction patterns
• Pipeline DSL: YAML syntax, node types, templates

Command Reference

Semantic Search (Primary)

# Vector-based semantic search
semantic-search <query> [--types Lifelog,Chat,Person,Topic] [--limit N] [--threshold 0.15] [--scores] [--json]

# Hybrid search (semantic + full-text)
search <query> [--mode semantic|fulltext|hybrid] [--types ...] [--limit N] [--json]

# Index management
index status                    # Show node counts, embeddings, vector indexes
index build --export-path <path> # Build from Limitless export

Graph Database

graph query <cypher> [--json]   # Execute Cypher query
graph stats                     # Show database statistics
graph traverse <label> <id>     # Traverse from node

Lifelogs

lifelogs list [--date YYYY-MM-DD] [--starred] [--limit N] [--json]
lifelogs get <id> [--json]
lifelogs search <query> [--limit N] [--json]

Pipelines

pipeline run <template|file> [--var key=value]
pipeline list       # List available templates

Available templates:

• daily-digest.yaml - Daily lifelog summary
• weekly-review.yaml - Comprehensive weekly analysis
• hierarchical-extraction.yaml - Full entity extraction
• session-extraction.yaml - Session detection
• extract-actions.yaml - Action item extraction
• memory-query.yaml - Context-augmented response
• research.yaml - Lifelog + web synthesis

Workflows

workflow daily <date>    # Complete day snapshot
workflow search <query>  # Cross-source search
workflow recent          # Recent activity summary

Domain Extraction Pattern

For domain-specific extraction, use this configurable pattern:

// Define domain patterns
const patterns = [
  { pattern: /ondansetron/gi, category: '5HT3 antagonist', displayName: 'Ondansetron' },
  { pattern: /droperidol/gi, category: 'Dopamine antagonist', displayName: 'Droperidol' },
];

// Filter lifelogs by domain keywords
const filtered = lifelogs.filter(l =>
  patterns.some(p => p.pattern.test(l.markdown))
);

// Extract using rule-based extraction
const results = extractFromLifelog(lifelog);

// Sync to graph with graceful degradation
try {
  await dbClient.connect();
  await lifelogRepo.upsert(lifelog);
} catch (error) {
  console.log(`⚠️ Graph sync skipped: ${error.message}`);
}

See references/extraction-rules.md for the full pattern library.

Environment Variables

LIMITLESS_API_KEY       # Required - API authentication
ANTHROPIC_API_KEY       # Optional - For LLM extraction
FALKORDB_HOST          # Default: localhost
FALKORDB_PORT          # Default: 6379

Graceful Degradation

The CLI handles missing dependencies gracefully:

Troubleshooting

"FalkorDBLite service not running"

• The Python service auto-starts on first use
• Manual start: cd ~/Projects/limitless-cli/python && uv run python -m limitless_graph.server
• Check socket: ls ~/.limitless/falkordb.sock

"No results from semantic search"

• Lower threshold: --threshold 0.1 (BGE scores are typically 0.15-0.35)
• Check embeddings: limitless index status
• Try different node types: --types Lifelog,Chat,Person,Topic

"Rate limited (429)"

• Wait for rate limit window (180 req/min, refills 3/sec)
• Check: config show for current settings

Integration with Other Skills

This skill integrates with:

• context-orchestrator: Use /limitless command for personal memory context
• error-recovery: Graceful degradation patterns applied
• deep-research: Combine with web research for synthesis pipelines

Project Location

~/Projects/limitless-cli/
├── src/
│   ├── api/           # Rate-limited API client (180 req/min)
│   ├── db/            # FalkorDB repositories (9 repos)
│   ├── dag/           # Pipeline engine (YAML DSL)
│   ├── extraction/    # Rule + LLM extraction
│   ├── agent/         # Claude Agent SDK harness
│   └── cli/           # Commander.js commands
├── templates/         # 7 pipeline templates
└── scripts/           # Demo and utility scripts