This is my implementation for pi-fff, for the official version make sure to checkout https://github.com/dmtrKovalenko/fff.nvim.

pi-fff

Screen.Recording.2026-04-09.at.12.20.45.PM.mov

pi-fff is a pi extension that brings fast fuzzy file finding and indexed content search into the editor, built-in tools, and agent workflows.

It is powered by @ff-labs/fff-node and gives pi a much better project-navigation experience.

What problem it solves

When working in a real codebase, agents and humans rarely know the exact path they need.

Instead of requiring exact file paths or repeated shell exploration, pi-fff lets pi:

• autocomplete fuzzy @... file references in the editor
• resolve approximate paths before read
• search code content with FFF-backed grep
• expose dedicated search tools for agents
• surface better ranked results and read suggestions

In short: it makes “find the right file fast” a first-class part of pi.

What the package adds to pi

pi-fff integrates at four levels.

1. Editor autocomplete

Type things like:

@readme
@src/index
@"folder with spaces/file"

pi-fff will offer ranked file suggestions from the current project.

2. Built-in tool upgrades

The extension can override pi’s built-in tools to make them FFF-aware. The read and grep upgrades are separate feature flags in /fff-features, so you can enable one without the other.

read

• accepts approximate file paths
• resolves fuzzy paths before reading
• supports @path-style input
• uses location hints from FFF when available

grep

• uses FFF-backed indexed content search
• supports fuzzy path or folder scopes
• pushes scope and glob filters down into native FFF constraints when possible
• returns more agent-friendly results
• falls back to pi’s built-in grep for compatibility edge cases

3. Custom agent tools

The extension also registers explicit tools for exploration and search:

• find_files
• resolve_file
• related_files
• fff_grep
• fff_multi_grep

These are useful when the agent wants to browse candidates first instead of immediately reading one guessed file.

4. Status and maintenance commands

Slash commands are included for feature flags, status, and manual recovery.

• /fff-features
• /fff-status
• /reindex-fff

How search and indexing work

pi-fff creates an FFF runtime for the current pi session.

That runtime:

1. indexes the current project on startup
2. keeps a background file watcher running
3. updates search results automatically as the project changes

So in normal use, indexing is automatic.

Do I need to reindex manually?

Usually, no.

/reindex-fff is mainly a fallback for cases like:

• very large branch switches
• mass file renames
• generated file churn
• stale results after watcher hiccups

Think of it as “force a fresh rescan if the index looks stale,” not something you should need after every edit.

What the agent gets

When active, pi-fff helps the agent work more efficiently by making it easier to:

• discover files related to a topic
• turn vague file references into exact paths
• search code with better ranking and constraints
• move from search results to a likely file to read next
• continue paginated searches with cursors

This reduces blind bash exploration and makes tool use more deliberate.

Search behavior highlights

Ranked file resolution

A vague query like:

editor

can resolve to the best matching file, or return ranked candidates if the result is ambiguous.

Better grep output

FFF grep output can include:

• a suggested file to read next
• definition-like matches near the top
• expanded context for likely definitions
• cursors for pagination

Multi-pattern content search

Use fff_multi_grep when you want to search for multiple literal variants at once, for example renamed symbols or aliases.

Commands

/fff-features

Interactive feature toggle UI.

• Space toggles the selected row
• Enter saves
• Esc cancels
• Built-in read and grep enhancements can be toggled independently
• State is stored globally in:
  • ~/.pi/agent/extensions/pi-fff.json

/fff-status

Shows:

• runtime state
• indexed file count
• repository detection
• current feature flags

/reindex-fff

Forces a rescan of the current project if the automatic watcher missed something.

Installation

From npm

pi install npm:pi-fff

From git

pi install git+https://github.com/ShpetimA/pi-fff.git

Restart pi after installation.

Development

npm install
npm run typecheck
npm test

Package metadata

{
  "pi": {
    "extensions": ["./index.ts"]
  }
}

Summary

pi-fff makes pi better at file discovery, fuzzy path resolution, and indexed code search.

If pi is active inside a repo, pi-fff helps both the editor and the agent get to the right file faster.

See PLAN.md for implementation notes and next steps.