mega-index-map

Working with DeepSeek Harness, every new conversation feels like starting from scratch: the directory structure mapped last session, the environment variables configured yesterday, the file formats already identified - all of it must be rediscovered by the agent before real work begins. mega-index-map solves this by persisting what agents learn into a local, cross-workspace memory library, so any subsequent conversation can index, search, and reuse prior discoveries without redundant exploration.

The tooling spans the full local-knowledge pipeline: recording objects with change detection (library_record), rebuilding and deduplicating the index (library_index), keyword and tag search with cursor pagination (library_query), scanning this machine for tools and environments with PATH-based resolution (library_detect), identifying files by 108 header-byte signatures regardless of extension (library_sniff), extending the local format library (library_format), decrypting isolated sensitive objects on request (library_decrypt), exporting to JSON or NDJSON (library_export), reporting host encoding (library_encoding), and driving Android devices over ADB (library_adb).

It is built for developers who juggle multiple projects in DeepSeek Harness and need environment and tool knowledge to follow them across workspaces. Everything runs locally with zero network I/O, works on Windows, macOS, and Linux, and stays fast - a query over 5,000 objects completes in roughly 13 milliseconds.