Firecrawl Open-source Autonomous Web Proxy Framework: Supports Multiple Models and Parallel Sub-proxies, One-click Generation of Research Tools

robot
Abstract generation in progress
ME News Report, April 17 (UTC+8), according to Beating Monitoring, Firecrawl officially open-sourced the Web Agent web proxy framework, aimed at helping developers build AI agents capable of autonomously executing "search - crawl - interaction" closed-loop tasks. The framework is based on Firecrawl's existing /agent architecture, supporting integration with Anthropic, OpenAI, or various self-hosted large models, emphasizing flexibility and self-hosting capabilities in web research scenarios. The core architecture of the framework adopts a "Plan-Act" cycle mechanism. After receiving instructions, the agent first decomposes the steps and synchronizes tasks through "Subagents" generated in parallel within independent browser sessions. This design has clear advantages when handling large-scale concurrent tasks, such as in video demonstrations where the agent can extract real-time stock data and news from multiple company websites simultaneously. For developers, this tool significantly lowers the barrier to developing web proxies. By using the command firecrawl create agent, a complete proxy template based on Next.js or Express can be quickly generated. Additionally, the framework introduces reusable "Skill Manuals" (SKILL.md), allowing developers to encapsulate complex workflows (such as specific crawling logic for Yahoo Finance) as Skills, which the agent can invoke directly in subsequent tasks. In the latest version, Web Agent has upgraded the default driving model to Claude Opus 4.7 and includes a specialized skill package for the financial sector. The project is currently open-sourced under the MIT license, encouraging community contributions through Forks. (Source: BlockBeats)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 9
  • 10
  • Share
Comment
Add a comment
Add a comment
GlassCityAfterTheRain
· 7h ago
The Plan-Act cycle sounds like a variant of ReAct, but its actual robustness depends on performance on long tasks.
View OriginalReply0
ReflectionsOnTheStreetAfterThe
· 14h ago
Community fork secondary development is the key point; let's see if anyone creates an e-commerce version of the Skill package.
View OriginalReply0
CatPawTapToConfirm
· 14h ago
Compatibility with self-hosted large models is very important; deploying on internal enterprise networks no longer depends on API availability.
View OriginalReply0
Frictionless
· 14h ago
Search - Crawl - Interaction Loop, finally no more bouncing back and forth between Playwright and Selenium.
View OriginalReply0
ProofOfVibes
· 14h ago
Claude Opus 4.7 default binding, Anthropic's integration here is a bit deep.
View OriginalReply0
GateUser-34d2b0ab
· 14h ago
Parallel sub-agent + independent browser session, this architecture is clearly designed for high concurrency scenarios.
View OriginalReply0
MoonlightMineralWater
· 14h ago
firecrawl create agent The command line experience is good, and the template generation speed is actually quite fast.
View OriginalReply0
SolitaryLampInTheSilentSea
· 14h ago
MIT License is highly praised, and the financial skill kit is ready to use out of the box, saving the trouble of building it from scratch.
View OriginalReply0
LookingAtTheCandlestickChart
· 14h ago
Finally, there is a decent open-source Agent framework, and the Plan-Act cycle design is quite clever.
View OriginalReply0
View More
  • Pinned