Roampal

Roampal is a memory system designed for AI coding assistants like Claude Code and OpenCode. It tracks which advice actually helped by recording user feedback—marking outcomes as successful or unsuccessful—and uses this data to improve future suggestions. The system stores memories locally and extracts summaries and atomic facts from conversations, organizing them with reliability scores based on past performance.

The tool uses a two-lane retrieval approach combining semantic search with outcome verification, achieving 85.8% accuracy on benchmark tests and outperforming standard RAG by 23 percentage points. It runs 100% offline with Ollama or LM Studio, is open source under Apache 2.0, and requires no signup or telemetry. The maker built it using Cursor and Claude over nine months.