Give your agents up to date knowledge on any api, framework, or repo

Shell access to a versioned filesystem of every doc, repo, and API spec — so your agents ground in current reality, not training cutoff.

$curl -fsSL https://fugg.dev/install | bash

FUGGbench Leaderboard

Every entry is a curated, versioned snapshot agents can shell into. FUGGbench scores each corpus on how well it can be navigated, parsed, and grounded by an LLM — not just whether it exists.

preview

Site	Score	Struct	Fresh	Version	Cold Nav	Scored
SwiftUI	92	96	95	92	84	2h ago
Stripe	90	95	95	90	80	4h ago
Tailwind CSS	88	92	95	84	86	6h ago
OpenAI API	86	88	95	78	82	3h ago
Next.js	85	90	95	86	70	4h ago
Expo	82	85	95	80	78	5h ago
Cobra	78	80	88	88	70	7h ago
LangChain	68	60	90	50	64	8h ago

Watch an agent use fugg

Five steps. Each one shows the verbatim fugg command the agent ran and the response it got back. Click through to see how the agent gets from a vague task to grounded code.

goal

Add a request rewriter to a Next.js 16 app — agent confirms the file convention before writing code, instead of guessing from training data.

What does fugg even have indexed?

Agent first asks fugg to list its corpora. Confirms vercel/nextjs is indexed and freshly synced.

command

$ fugg sites

response

apple/activitykit
apple/appintents
apple/appstoreconnectapi
apple/charts
apple/cloudkit
apple/foundation
apple/swiftui
expo/docs
langchain/python
openai/api
spf13/cobra
stripe/docs
tailwindcss/docs
vercel/nextjs
… 25 more

what this gives your agent

Speed

Lightning fast

Indexes documentation in seconds. Zero latency for agent queries — your agent reads from a versioned filesystem, not a remote API.

Accuracy

Grounded search

Combines exact keyword matching with structural awareness to find exactly what your agents need, every time.

Integration

Agent ready

Outputs clean, structured JSON or Markdown specifically formatted for LLM context windows. Works with any agent that can shell out.

About

fugg is an open source agent tool that gives your agent shell access to a versioned filesystem of every doc, repo, and API spec it might need — so it grounds in current reality, not its training cutoff.

Agent-first Built specifically for AI agents to consume documentation
Grounded search Natural-shell queries with ranked, relevant results
Multi-source Index from GitHub repos, wikis, API refs, local markdown
Structured output JSON, Markdown, or context-window-optimized format
Zero config Works out of the box with sensible defaults