ContextSlim 🧠✂️

Turn messy websites into high-signal, "Agent-Ready" knowledge bases while cutting your LLM costs by up to 90%.

ContextSlim is not just another web crawler. It is a specialized Knowledge Architect designed for the era of AI Agents. While standard scrapers provide "data dumps" full of navbars, footers, and marketing fluff, ContextSlim uses structural heuristics to strip away the noise and deliver pure, semantically organized context.

---

🚀 Why ContextSlim?

AI Agents (Claude, GPT-5, etc.) are only as good as the context you give them. However, feeding them raw scraped HTML or uncurated Markdown leads to:

1. Token Bloat: You pay for "Home," "Login," and "Copyright 2026" over and over.
2. Hallucinations: Agents lose key facts in a sea of irrelevant links.
3. Context Window Exhaustion: Massive sites won't fit into a single prompt.

The Cost-Saving Formula

For every crawl, ContextSlim calculates your savings:

Savings = (Original Tokens - ContextSlim Tokens) × Cost per Token

Where N_tokens is the number of tokens and C_token is the cost per token of your chosen LLM.

---

✨ Key Features

• The "Noise-Killer" Engine: Advanced filtering that identifies and removes <nav>, <footer>, <aside>, and <form> elements before conversion.
• Semantic Knowledge Bricks: Automatically categorizes content into logical sections (e.g., ## Pricing, ## Documentation, ## Technical Specs).
• Agent-Optimized Export:
  • Claude Project Text: Perfectly formatted for the "Add Content" button in Claude.
  • MCP Ready: JSON output structured for Model Context Protocol servers.
  • GPT Knowledge: Cleaned Markdown for Custom GPT uploads.
• Smart Link Pruning: Avoids scraping "Legal," "Privacy Policy," and "Terms of Service" unless explicitly requested.

---

🛠 How It Works (The Technical Edge)

ContextSlim uses the Apify PlaywrightCrawler to render JavaScript-heavy sites, then applies a custom DOM-Purification Layer:

1. Rendering: Executes JS to ensure data hidden behind tabs or toggles is captured.
2. Heuristic Analysis: Analyzes tag density and link-to-text ratios to distinguish between "Navigation" and "Content."
3. Markdown Distillation: Converts the purified HTML into clean, hierarchical Markdown.
4. Token Estimation: Provides a real-time count of tokens saved compared to a standard crawl.

---

💰 Monetization (Pay-Per-Event)

This Actor uses the Apify PPE (Pay-per-Event) model to provide maximum value for a low entry price:

• $0.10 per "Knowledge Architecture Event"
• An event includes crawling and architecting up to 50 pages from a single domain.

"Why spend $5 on a general scraper and $2 in LLM tokens to read the mess? Pay $0.10 for ContextSlim and get the 5% of the data that actually matters."

---

📥 Input Schema

Field	Type	Description
startUrls	Array	The entry point for the crawl.
maxDepth	Integer	How many clicks deep the architect should go (Default: 2).
exportFormats	Array	md, txt, mcp (Default: all).
smartPrune	Boolean	Toggle the "Noise-Killer" engine (Default: True).
costPer1kTokens	Number	Used for the shrink report estimate (Default: 0.01).

---

🏁 Get Started

1. Add your Start URL.
2. Select your Export Formats (e.g., "md", "mcp").
3. Run the Actor.
4. Click Storage after agent is done
5. Your data is under Key Value Storage
6. Download the files :
   - openai_knowledge.md for ChatGPT agents
   - knowledge_bricks.json for Claude agents
   - mcp_schema.jsom for setting up MCP servers
7. Now you're free to use them to train your agent on the documentation 😁

---

Built for the $1 Hackathon: The First Dollar Sprint by AI.SEA. Solving the "Token Bloat" problem, one page at a time.