MemoryLane 🧠

Note: "MemoryLane" is a working title. The name may change before v1.0. Name suggestions welcome!
Local-first persistent memory for Claude Code with automatic cost savings
MemoryLane gives Claude Code durable project memory by extracting knowledge locally and injecting only what matters. It reduces token usage by up to 84%, eliminates repeated explanations, and keeps raw development sessions on your machine.
⚠️ Alpha Software
This is a proof of concept and alpha-stage software.
- Not tested in production environments
- Memory quality and cost savings claims need real-world verification
- API and configuration may change without notice
- Use at your own risk
We welcome feedback, bug reports, and contributions to help mature this project.
🎯 Core Value
Turn raw dev sessions into durable knowledge.
Only intentional, compressed context is sent to the cloud.
✨ Features
-
🔄 Zero Configuration
One-command install. Automatic learning and context injection.
-
💰 Massive Cost Savings
30–84% token reduction through intelligent summarization and compression.
-
🧠 Claude-Powered Extraction
Uses Claude (via CLI or API) with trigger-specific prompts to extract design decisions, solutions, patterns, and architectural insights.
Falls back to local LLM or regex heuristics when Claude is unavailable.
-
🔒 Privacy-First by Design
Raw transcripts, diffs, and files stay local. Only curated context reaches Claude.
-
📊 Passive Learning
Learns from file edits, git commits, and CLI sessions without workflow changes.
-
⚡ Fast
<100ms recall latency. Async summarization (<5s typical).
-
🧯 Context Rot Guard
Automatically caps injected context to a safe fraction of the model’s window to prevent quality decay at long context lengths.
-
📈 Cost Tracking
Transparent token savings with validated test coverage.
🚀 Quick Start
Installation
# Install the Claude Code skill from GitHub
claude skill install AWolf81/memory-lane --skill memorylane
# Or install via marketplace alias (one-time add, then install)
claude plugin marketplace add AWolf81/memory-lane
claude plugin install memorylane@memorylane
# Or manual installation
git clone https://github.com/AWolf81/memory-lane.git
cd memorylane
bash install.sh
Usage
# Check status and cost savings
python3 src/cli.py status
# Recall memories about a topic
python3 src/cli.py recall "authentication"
# View learned insights
python3 src/cli.py insights
# See detailed cost breakdown
python3 src/cli.py costs
That’s it. MemoryLane now learns from your project and injects context automatically.
📊 Validated Cost Savings
Cost savings are measured, not estimated.
pytest tests/test_cost_savings.py -v -s
Realistic Weekly Usage Test
- 100 Claude interactions/week
- Baseline: ~2.3M tokens/week
- MemoryLane: ~360K tokens/week
- Compression: 6.4×
- Savings: 84.3%
- Monthly: ~$51 saved per developer
🏗️ Architecture (High-Level)
MemoryLane is local-first and session-aware: