fff.nvim: The File Search That Saves Your Agent 10x on Tokens
The biggest hidden cost of coding agents isn't the LLM call — it's the file search. Every time an agent needs to find a function definition, it's doing round trips: glob, read, grep, read again. Each round trip costs tokens and time. fff.nvim attacks this problem at the root.
Built in Rust with a Lua frontend for Neovim, fff maintains a separate file index with a dedicated backend process. Result: under 10 milliseconds search time for a 50,000-file codebase. For humans, it provides typo-resistant fuzzy finding. For AI agents, it provides the fastest file search available, with built-in memory that ranks results by frecency, git status, file size, and definition matches.
The real value for agents is the MCP server binary that ships with the project. Plug it into Claude Code, Codex, or any MCP-compatible agent and your file search gets faster while spending fewer tokens. The agent does fewer roundtrips and reads fewer useless files. That directly translates to lower cost and faster task completion.
Trending on GitHub today at 767 stars per day, 3,200 total. The repo is at github.com/dmtrKovalenko/fff.nvim. Most agent tools focus on making the model smarter. fff.nvim focuses on making the model's eyes faster. Sometimes the bottleneck isn't intelligence — it's I/O.
https://github.com/dmtrKovalenko/fff.nvim
← Back to all articles
Built in Rust with a Lua frontend for Neovim, fff maintains a separate file index with a dedicated backend process. Result: under 10 milliseconds search time for a 50,000-file codebase. For humans, it provides typo-resistant fuzzy finding. For AI agents, it provides the fastest file search available, with built-in memory that ranks results by frecency, git status, file size, and definition matches.
The real value for agents is the MCP server binary that ships with the project. Plug it into Claude Code, Codex, or any MCP-compatible agent and your file search gets faster while spending fewer tokens. The agent does fewer roundtrips and reads fewer useless files. That directly translates to lower cost and faster task completion.
Trending on GitHub today at 767 stars per day, 3,200 total. The repo is at github.com/dmtrKovalenko/fff.nvim. Most agent tools focus on making the model smarter. fff.nvim focuses on making the model's eyes faster. Sometimes the bottleneck isn't intelligence — it's I/O.
https://github.com/dmtrKovalenko/fff.nvim
Comments