SpeedSearch — Fast local search bridge for coding agents

Let AI coding assistants search local files at Everything speed instead of crawling with grep

Problem

One developer shipped a skill that makes AI search local files with Everything, arguing that even at hundreds of tokens per second, the model still kneels before file search. Meanwhile another hands-on test showed LSP-based semantic navigation is not clearly better than grep: in simple locate tasks the model picked LSP only 0-6% of the time, and forcing it dropped success from 100% to 89%. There's no default search setup people are happy with.

Solution

A cross-platform local index service that wraps native search (Everything on Windows, Spotlight on macOS, locate on Linux) into a single interface agents can call. Agents send a pattern or natural-language query and get ranked file paths with snippets, plus a hint on whether full-text search or semantic jump fits better, cutting unnecessary large-file reads.

Users

Developers using Claude Code, Codex and similar coding agents, especially Windows users and large-repo maintainers.

MVP

Windows-only at first, talking to Everything's HTTP/IPC API; ship a standard MCP server; support wildcards, regex and path filters; return structured hit lists; include a token-spend comparison log.

Revenue

Open-source core; annual Pro subscription adds multi-machine index sync, query caching and priority support.

Why now

Coding agents are becoming the default developer entry point, file search is one of their most frequent and expensive actions, and current options are either slow or blindly chose the wrong tool.