LogoAIRosetta

OpenCLI Emerges: the "Everything as CLI" Era Begins as AI Agents Move Beyond GUI

WeChat

SummaryOpen-source OpenCLI uses a Chrome extension and a local daemon to turn websites and Electron apps into CLI tools callable by AI Agents. Paired with CLI-Anything for desktop software, it enables scenario-level automation and structured output but raises browser session permission and security risks.

On March 20, 2026, an open‑source project named OpenCLI quietly went viral on GitHub, gaining over 2,300 stars within days. Led by developer jackwener, its core idea is to "turn any website, Electron app, or desktop software into a CLI tool," allowing AI Agents to stop awkwardly "clicking buttons and filling forms" and instead perform cross‑platform, structured operations directly via the command line. At the same time, a parallel project, CLI‑Anything (HKUDS team), has accumulated 15,000 stars. The two projects complement each other ⁠—⁠ one starts from the browser/web, the other from open‑source desktop application source code ⁠—⁠ jointly announcing the start of an Agent era that claims "GUI is dying, CLI is everything."

This wave of "CLI‑ification" is not an isolated event but an inevitable outcome of AI Agent technology evolving from an "assistive tool" to "autonomous execution." Karpathy once bluntly said: "Making users click buttons and fill forms on web pages feels crude." Levie put it even more directly: "If your feature has no API, it effectively doesn't exist." OpenCLI is a practical realization of these insights: it gives AI Agents a true "machine language" ⁠—⁠ the CLI ⁠—⁠ enabling a fundamental reconnection of the software world.

Background: Forty Years of GUI "Cosmetics" Are Finally Coming Off

For the past 40 years, humans have wrapped computers in a thick GUI layer: buttons, forms, menus, drag‑and‑drop. These made computing accessible to people but became the biggest obstacle for AI Agents. GUIs are a "human language": inputs are pixel‑level clicks, outputs are visual feedback ⁠—⁠ neither composable, discoverable, nor deterministic. CLI is the exact opposite: text input, structured output, composable, discoverable (‑‑help), and highly deterministic ⁠—⁠ the ideal interaction protocol for the Agent era.

Previously, if an Agent wanted to operate on siloed services like Bilibili, Zhihu, Xiaohongshu, Feishu, Notion, etc., it had to rely on three inefficient methods: manual copy‑paste, third‑party unofficial APIs (often unreliable), or web scraping (very unstable). Large companies use private APIs for cross‑system orchestration, but ordinary developers and personal Agents had no practical access.

The arrival of OpenCLI and CLI‑Anything breaks that barrier. The two projects have clear roles:

  • CLI‑Anything: Targets open‑source desktop software (GIMP, Blender, Audacity, LibreOffice, etc.), scanning source code and automatically mapping GUI operations into Python Click‑based CLI commands.
  • OpenCLI: Targets websites and Electron applications (covering nearly half of modern desktop apps) and requires no source code, operating directly through browser sessions.

Together they realize the vision that "anything can be CLI, and anything should be CLI."

OpenCLI Technical Breakdown: Browser Session Reuse + Three‑Step Automation

OpenCLI's core architecture is extremely lightweight and consists of two parts:

  1. Lightweight Chrome extension (Browser Bridge): After installation, it reuses the user's logged‑in Chrome session seamlessly with no additional sign‑in; passwords and credentials remain in the browser.
  2. Local daemon: Communicates with the extension via WebSocket and listens on localhost:19825. It auto‑exits after 5 minutes of idleness and has minimal resource usage.

The full invocation chain is:

CLI command → local daemon → WebSocket → Chrome extension → webpage actions.

The project includes three core commands that enable zero‑code automation for Agents:

  • explore: Given a URL, it automatically opens the browser, clicks, scrolls, observes network requests, and dynamically discovers callable APIs.
  • synthesize: Based on explore results, it automatically generates a CLI adapter (no manual coding required), packaging the API as a standard CLI tool.
  • cascade: Intelligently detects authentication strategies (five levels), starting from public APIs and progressively escalating to cookie‑based auth and request‑signature interception, maximizing compatibility with different sites.

One command can complete the whole process:

opencli generate https://bilibili.com ‑‑goal "hot"

This generates a Bilibili "hot list" CLI tool; afterwards you can run opencli bilibili hot -f json to output structured data for further Agent processing.

OpenCLI currently includes 80+ commands covering 30+ mainstream sites and apps:

  • Video platforms: Bilibili (hot list, search, subtitle extraction), YouTube (video info)
  • Communities: Zhihu (hot list, search), Xiaohongshu (note download), X/Twitter (timeline, bookmarks), Reddit (hot posts)
  • Tools: Xueqiu (stock data), Boss Zhipin (job search), Feishu, Notion, Discord, NetEase Cloud Music, Chaoxing Learning, WeRead, etc.
  • Electron apps: Cursor, ChatGPT clients, Notion, Figma, VS Code, Slack, WeChat Developer Tools, etc. (using Chromium‑based apps with fixed ports, e.g. Cursor 9222, Notion 9230)

For complex scenarios such as React rich‑text editors, OpenCLI cleverly uses document.execCommand('insertText') to simulate input, effectively bypassing framework state‑update issues.

Perfect Complement to CLI‑Anything: Covering the Entire Software Ecosystem

CLI‑Anything focuses on open‑source desktop software and already supports dozens of tools like GIMP and Blender by generating CLIs through source analysis. OpenCLI specializes in the web and Electron apps (nearly half of modern desktop apps are Chromium‑based). Combined, they make it possible for "any software, any codebase, any Web API" to become Agent‑native.

Developers can install via a single pip command or GitHub Marketplace action, enabling AI Agents (OpenClaw, Cursor, Claude Code, etc.) to call them directly. Structured outputs support JSON, YAML, Markdown, and CSV, fitting downstream Agent workflows.

Industry Impact: Exponential Growth in Agent Capabilities

OpenCLI directly addresses the biggest pain point of the Agent era ⁠—⁠ "software silos." Previously, cross‑platform tasks (for example, extracting subtitles from Bilibili, having AI summarize them, then posting to a Feishu group) required manual intervention; now a single CLI command can fully automate the process.

For individual developers, this means dramatically higher solo‑team productivity; for enterprises, it reduces internal orchestration costs; for the AI ecosystem, the CLI may become the standard protocol for Agents while GUIs gradually return to being "human‑only" interfaces.

GitHub star counts most directly reflect interest: CLI‑Anything has reached 15,000 stars, and OpenCLI surpassed 2,300 stars within days of release. Community feedback has been enthusiastic, calling this a key step toward enabling Agents to "do things on their own."

Potential Risks and Future Discussion

Powerful capabilities come with responsibility. Because OpenCLI reuses a user's browser session, an Agent's permissions can match the user's permissions ⁠—⁠ it can post, delete, transfer funds, or download sensitive files. This raises the core question: "Who decides what an Agent may or may not do?"

Developers should be cautious:

  • Use dedicated browser profiles for isolation
  • Enforce strict permission controls
  • Avoid using sensitive accounts with Agents

The future direction is clear: more platforms will natively support CLI adapters; OpenCLI and CLI‑Anything may merge into a unified hub; Agents will evolve from "being able to chat" to "being able to act," and the software world will be thoroughly reconnected.

Conclusion: CLI Is the Native Language of the Agent Era

From Karpathy's "clicking buttons is crude," to Levie's "no API equals no existence," to the practical implementations of OpenCLI and CLI‑Anything, the 40‑year GUI layer is being systematically removed.

OpenCLI is not a simple tool but a paradigm shift in software interaction. It gives AI Agents their first true "native language" ⁠—⁠ the CLI ⁠—⁠ enabling cross‑platform, structured, automated execution.

In the 2026 Agent wave, "everything can be CLI" is no longer a slogan but a reality in progress. The entire software world is being reconnected and plugged back in. We stand at the forefront of this revolution.

Sources:

1. OpenCLI GitHub repository (https://github.com/jackwener/opencli)

2. CLI‑Anything GitHub repository (https://github.com/HKUDS/CLI‑Anything)

3. Author J0hn's WeChat article "OpenCLI: Everything Can Be CLI" (March 20, 2026)

▶️ Enjoy this article? Creating content is hard work!

Follow our YouTube channel:
AI Rosetta

Your likes, subscriptions, and shares are our greatest support! 🙏

More AI news at: https://www.airosetta.com

CopyrightStatement

Comment

Please login before commenting

NoComments

RelatedNews

LoadingMore

OpenCLI Emerges: the "Everything as CLI" Era Begins as AI Agents Move Beyond GUI - Airosetta News