🌐 Tool: @fetch

The @fetch tool allows the AI agent to pull specific web pages directly into the context window, bypassing the agent's limited training data bounds and ensuring context perfectly aligns with upstream documentation.

Technical Implementation

Implemented in crates/agent/src/tools/fetch_tool.rs, the tool utilizes a headless Chromium instance (via the Chromiumoxide library).

Unlike standard HTTP libraries, @fetch ensures:

• Client-Side Rendering: Executes JavaScript (React, Vue, etc.) to capture content on modern single-page applications.
• Markdown Transformation: Automatically converts the resulting DOM into clean, hierarchical Markdown.

Usage Examples

Example 1: Updating out-of-date library syntax

User

Update my `Cargo.toml` and main file to use the latest `axum` patterns.

Zed Agent

The model has a training cutoff, but I can fetch the live installation docs.

Tool Output

@fetch({ url: "https://docs.rs/axum/latest/axum/" }) >> Status: Rendered (JS Hydrated) >> Using latest 0.7.x patterns...

Example 2: Reading API Specifications

User

What are the valid headers for the Anthropic beta prompt caching API?

Zed Agent

Fetching Anthropic's developer guide...

Tool Output

@fetch({ url: "https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching" }) >> Extracting `anthropic-beta: prompt-caching-2024-07-31` header requirements.

Key Advantages & Impact

• Dynamic Context: Allows the agent to use libraries that were released after its knowledge cutoff.
• Accurate Syntax: By reading the official docs, the agent avoids deprecated API patterns and solves "it works on my machine" issues.
• Privacy: All rendering happens within a local Chromium process; no third-party scraping services are used.
• Bypass Paywalls/Restrictions: Handles complex login-less modern docs that simple curl-based scrapers can't see.