GroundTruth

Real-time documentation injection for AI coding agents. No config. No API keys. Just run.

---

The Problem

AI coding assistants like Claude Code, Cursor, and Gemini have a knowledge cutoff. When you're building with Svelte 5, React 19, or any fast-moving framework, your agent is coding blind — working from stale docs that may be months or years out of date.

The Solution

GroundTruth is a transparent middleware layer that fetches live documentation and injects it directly into your agent's context window before every inference. Zero config. Zero API keys. One command.

---

Quick Start

# For Claude Code (proxy mode)
npx @antodevs/groundtruth@latest --claude-code

# For Gemini / Antigravity (file watcher mode)
npx @antodevs/groundtruth@latest --antigravity

That's it. GroundTruth runs in the background and handles everything automatically.

---

How It Works

GroundTruth operates in two modes depending on your agent:

1. Proxy Mode — for Claude Code

Spins up a local HTTP proxy that intercepts every outbound API call to Anthropic, enriches the system prompt with fresh documentation, then forwards the mutated request.

npx @antodevs/groundtruth@latest --claude-code
# Then set: ANTHROPIC_BASE_URL=http://localhost:8080

sequenceDiagram
    participant Agent as Claude Code
    participant Proxy as GroundTruth
    participant Jina as Jina Reader API
    participant API as Anthropic API

    Agent->>Proxy: Send Prompt
    Proxy->>Jina: Fetch docs (Registry / DDG)
    Jina-->>Proxy: Return clean Markdown
    Note over Proxy: Injects live context<br/>into System Prompt
    Proxy->>API: Forward mutated request
    API-->>Agent: Response with fresh context

Loading

2. Antigravity Mode — for Gemini / Antigravity

Runs as a background daemon that watches your package.json, fetches docs for your exact dependency versions, and writes them into ~/.gemini/GEMINI.md automatically.

npx @antodevs/groundtruth@latest --antigravity

flowchart TD
    pkg([package.json]) -->|Parse Dependencies| GT{GroundTruth Watcher}
    GT -->|Smart Routing| Map{Registry?}
    Map -->|Yes| Jina[Jina Reader API]
    Map -->|No| DDG[DuckDuckGo Search] --> Jina
    Jina -->|Clean Markdown| Gen[Write to ~/.gemini/GEMINI.md]
    Gen --> Agent(Coding Assistant)

Loading

---

The Engine

GroundTruth's context pipeline is built for precision and speed:

• Cloudflare Worker Registry — skips search entirely for known packages. Covers the top ~200 frameworks with hand-curated URLs and 10,000+ npm packages via automated indexing.
• Jina Reader API — renders JavaScript-heavy SPAs (Next.js, Svelte, Vercel AI SDK) into clean LLM-optimized Markdown.
• DuckDuckGo Fallback — for packages not in the registry, falls back to live web search with HTML parsing.
• Zero-config resilience — strict 1.5s timeout on the cloud registry. If anything fails, it silently falls back. Your agent never blocks.

---

Installation

Via npx (recommended — always latest)

npx @antodevs/groundtruth@latest --antigravity
npx @antodevs/groundtruth@latest --claude-code

Global install

npm install -g @antodevs/groundtruth
groundtruth --antigravity

---

Configuration

Create a .groundtruth.json in your project root for custom settings:

{
  "maxChars": 4000,
  "quality": "high",
  "verbose": true,
  "sources": [
    { "url": "https://svelte.dev/docs/kit/introduction", "label": "SvelteKit Docs" },
    { "url": "https://your-internal-wiki.com/api", "label": "Internal API" }
  ]
}

Option	Type	Default	Description
maxChars	number	4000	Max characters injected per context block
quality	low | medium | high	medium	Controls search depth and timeout budget
verbose	boolean	false	Enables detailed fetch logging
sources	array	[]	Custom URLs to always inject (internal docs, wikis, etc.)

Backward compatible: maxTokens is still accepted as an alias for maxChars.

---

CLI Reference

Flag	Mode	Description
--claude-code	Proxy	Starts HTTP proxy interceptor for Anthropic API
--antigravity	Watcher	Starts background daemon for dotfile injection
--uninstall	Cleanup	Removes ANTHROPIC_BASE_URL from all shell configs
--port <n>	Proxy	Proxy port (default: 8080)
--quality <level>	Both	low, medium, or high (default: medium)
--max-chars <n>	Both	Character limit per context block (default: 4000)
--interval <n>	Watcher	Refresh interval in minutes (default: 5)
--batch-size <n>	Watcher	Dependencies per search batch (default: 3)
--verbose	Both	Enable verbose logging

---

Comparison

Feature	GroundTruth	Jina Reader (direct)	Crawl4AI / Playwright	Firecrawl
Setup	✅ Zero (1 command)	⚠️ Scripting needed	❌ Docker + deps	❌ API key required
JS Rendering	✅ Via Jina	✅ Yes	✅ Yes	✅ Yes
Agent Injection	✅ Automatic	❌ Manual	❌ Manual	❌ Manual
Works offline	✅ Graceful fallback	❌ No	❌ No	❌ No
Cost	✅ Free	⚠️ Rate limits	✅ Free	❌ Paid
Runtime size	✅ < 1MB	—	❌ ~200MB	—

---

Requirements

• Node.js v18.0.0 or higher
• Claude Code, Antigravity, or any agent that supports system prompt injection or dotfile context

---

Contributing

Pull requests are welcome. For major changes, please open an issue first.

git clone https://github.com/anto0102/GroundTruth.git
cd GroundTruth
npm install
npm test        # run full test suite
npm run dev     # run with hot reload

---

License

MIT © Anto