claude-code-langcache

Semantic caching skill for Claude Code using Redis LangCache.

Reduce LLM costs and latency by caching responses for semantically similar queries, with built-in privacy and security guardrails.

For OpenClaw users: See openclaw-langcache

Features

• Semantic similarity matching - Cache hits for similar (not just identical) queries
• Hard block enforcement - Automatically blocks caching of sensitive data:
  • Temporal info (today, tomorrow, deadlines, appointments)
  • Credentials (API keys, passwords, tokens, OTP)
  • Identifiers (emails, phone numbers, account IDs)
  • Personal context (relationships, private conversations)
• Category-aware thresholds - Different similarity thresholds for factual Q&A vs style transforms
• CLI and Python integration - Use from shell scripts or embed in Python agents

Installation

Via npm (Recommended)

npm install -g claude-code-langcache

The skill will be automatically installed to ~/.claude/skills/langcache/

Via Git

git clone https://github.com/manvinder01/claude-code-langcache.git /tmp/claude-code-langcache
cp -r /tmp/claude-code-langcache/skills/langcache ~/.claude/skills/

Configuration

Set your Redis LangCache credentials:

export LANGCACHE_HOST=your-instance.redis.cloud
export LANGCACHE_CACHE_ID=your-cache-id
export LANGCACHE_API_KEY=your-api-key

Get these from Redis Cloud Console after creating a LangCache instance.

Usage

Automatic (via Claude Code)

The skill triggers automatically when you mention:

• "cache LLM responses"
• "semantic caching"
• "reduce API costs"
• "configure LangCache"

Or invoke manually with /langcache

CLI

# Search for cached response
~/.claude/skills/langcache/scripts/langcache.sh search "What is Redis?"

# With similarity threshold
~/.claude/skills/langcache/scripts/langcache.sh search "What is Redis?" --threshold 0.9

# Store a response
~/.claude/skills/langcache/scripts/langcache.sh store "What is Redis?" "Redis is an in-memory data store..."

# Check if content would be blocked
~/.claude/skills/langcache/scripts/langcache.sh check "What's on my calendar today?"
# Output: BLOCKED: temporal_info

Caching Policy

Cacheable (white-list)

| Category | Examples | Threshold | |----------|----------|-----------| | Factual Q&A | "What is X?", "How does Y work?" | 0.90 | | Definitions / docs | API docs, command help | 0.90 | | Command explanations | "What does git rebase do?" | 0.92 | | Reply templates | "polite no", "follow-up", "intro" | 0.88 | | Style transforms | "make this warmer/shorter" | 0.85 |

Never Cached (hard blocks)

| Category | Examples | |----------|----------| | Temporal | today, tomorrow, deadline, ETA, "in 20 minutes" | | Credentials | API keys, passwords, tokens, OTP/2FA | | Identifiers | emails, phone numbers, account IDs, UUIDs | | Personal | "my wife said", private conversations, relationships |

File Structure

~/.claude/skills/langcache/
├── SKILL.md              # Skill definition and instructions
├── scripts/
│   └── langcache.sh      # CLI wrapper with policy enforcement
├── references/
│   ├── api-reference.md  # Complete REST API documentation
│   └── best-practices.md # Optimization techniques
└── examples/
    ├── basic-caching.sh      # Simple cache workflow
    └── agent-integration.py  # Python integration pattern

Requirements

• Claude Code
• Redis Cloud account with LangCache enabled
• Node.js 18+ (for npm installation)
• jq and curl (for CLI usage)

Related Packages

• openclaw-langcache - For OpenClaw users

License

MIT License - see LICENSE for details.

Resources

• Redis LangCache Documentation
• Claude Code Documentation
• Semantic Caching Guide