memex-search

The memex-search skill provides a robust solution for indexing, searching, and retrieving historical logs from Claude and Codex interactions using the memex CLI. It addresses the common challenge of efficiently managing and querying large volumes of conversation and tool usage history, allowing users to quickly locate relevant information across multiple sessions. By supporting both lexical and semantic search modes, as well as hybrid approaches, this skill enables precise retrieval of data while maintaining structured outputs suitable for further automated processing or integration with language models for retrieval-augmented generation (RAG) workflows.

Key features of memex-search include incremental and full indexing of local history, embedding support for semantic queries, and continuous background indexing for keeping data up to date. Users can run targeted searches with JSON or JSONL output formats, filter results by project, role, tool, session, source, date range, and score, and group or deduplicate hits to focus on the most relevant records. Additional capabilities include selecting different embedding models, adjusting index root paths, and customizing the output fields. The skill also supports macOS background indexing via launchd and provides practical strategies for efficient query narrowing, ensuring users can handle large datasets without performance degradation.

This skill is particularly useful for developers, researchers, and data analysts who need to track, audit, or analyze interactions with AI systems over time. Use cases include exploring assistant responses, debugging tool outputs, analyzing project-specific sessions, or preparing structured datasets for downstream tasks such as training or fine-tuning language models. Target users are those who require fast, structured, and reproducible access to historical AI interactions, including software engineers, AI researchers, and technical documentation specialists.