> ## Content Index > Fetch the complete content index at: https://alohabrowser.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # AlohaJet: A Browser Built for AI Agents - URL: https://alohabrowser.ghost.io/posts-alohajet-browser-for-ai-agents/ - Published: 2026-09-29T07:47:20.000Z - Updated: 2026-09-29T08:36:33.000Z - Description: AlohaJet gives AI agents a browser for real websites, with a visible mode for oversight and headless automation for CLI and MCP workflows. - Author: Aloha Browser - Tags: Aloha What's New, AlohaJet Running a browser task usually requires minimal resources. However, performing the same task thousands of times can increase model usage and maintenance costs. The AlohaJet browser helps agents work with real websites more efficiently by reducing page processing time and minimizing repetitive workflow setup. AlohaJet is a browser tailored for agents who interact with real websites. It lets agents understand a page and take action via the command line or MCP, with the option to keep the browser visible or run it in the background. ## Give the agent a clearer view of the page AlohaJet uses LLMdex to give a model a compact understanding of a webpage and the specific task. This lets the model focus on relevant controls and information rather than sifting through the entire DOM or relying on screenshots at every step. LLMdex also routes each step to the most cost-effective model that can handle it. For repeated tasks, it can store and replay a route, reducing setup time for future runs. This setup simplifies workflow creation and maintenance for developers building web agents. AlohaJet suits practical tasks such as gathering prices or reviews, navigating forms and dashboards, and checking information across multiple sites. For teams performing the same workflow repeatedly, reducing token usage and setup effort can lower costs. The WebArena benchmark reports an 88% success rate for AlohaJet, compared to 51% for Chrome DevTools MCP and 11% for Playwright MCP. Average task times were 39 seconds for AlohaJet, 73 seconds for Chrome DevTools MCP, and 84 seconds for Playwright MCP. Aloha reports an average of 158,000 tokens per task for AlohaJet, versus 340,000 tokens for Chrome DevTools MCP; no token data is listed for Playwright. The results combine GPT-5-mini and Qwen3-30B across 10 repetitions per task. See the [benchmark setup and results](https://alohajet.com/benchmarks/). Results can vary by website and task. Note that outcomes may vary by website and task. ![Run web agents and save tokens for repeated work](https://storage.ghost.io/c/9b/e8/9be874bd-0377-40be-b16f-9c460166ff13/content/images/2026/09/Blog.-AlohaJet-1200x630-1-2.png) ## Watch a task or run it headlessly Some jobs are easier to debug with a visible browser. In headed mode, you can follow each step and take control when needed. For automated tasks, headless mode runs via the CLI or MCP without opening a browser window. The product site explains how to use it in pipelines, CI, and on servers. ## Choose the model that fits AlohaJet works with tool-calling models, including frontier model APIs and open-source models running locally. You can select a model for your task and connect to it via the CLI or the MCP setup. Developers can find the command-line and MCP browser-control tools in the public [AlohaJet CLI repository on GitHub](https://github.com/AlohaBrowser/alohajet-cli). The README explains the features, supported browser modes, and limitations. AlohaJet is designed to run directly on your device without an Aloha SaaS relay. If you connect to a hosted model, review the provider's data practices for your requests. Alternatively, you can use a local model to keep inference on your machine. ## Start with AlohaJet AlohaJet is available on macOS, and a Linux version is coming soon. The service offers a free plan, along with paid options for additional features. Visit the [download](https://alohajet.com/download/) and [pricing](https://alohajet.com/pricing/) pages for the latest availability and plan details. ## AlohaJet browser FAQ AlohaJet is a browser built for AI agents that need to interact with real websites. These answers cover how it works, which models and platforms it supports, and what to check before you start. ### How does AlohaJet help an AI agent work with websites? AlohaJet uses LLMdex to provide a concise understanding of a page and the task. It can also direct each step to a capable, affordable model and reuse saved routes for repeated tasks. ### Can I watch an agent work and take control? In headed mode, you can watch the browser as the agent works, debug visually, and take control when necessary. Headless mode runs without a visible browser window, suitable for unattended workflows. ### Which AI models work with AlohaJet? AlohaJet integrates with tool-calling models, including frontier model APIs and open-source models run locally. The connection process varies by your model and setup. For detailed instructions, see the [CLI and MCP quickstart](https://alohajet.com/headless/). ### Does AlohaJet use an Aloha cloud relay? No mandatory Aloha SaaS relay is required. AlohaJet operates on your device. If you connect a hosted model, the provider manages requests sent to it according to its terms and data practices. Please review the [AlohaJet privacy policy](https://alohajet.com/privacy/) for the latest information. ### Can I use AlohaJet for free? Yes. The AlohaJet site lists a free plan, along with paid Pro and Ultra plans. Plan limits and prices may change, so check the [pricing page](https://alohajet.com/pricing/) for current details. ### What can an agent do with AlohaJet? Examples include gathering prices, reviews, listings, or other structured data; completing multi-step tasks on various websites; and tracking information across multiple sites. The outcomes vary depending on the sites and how the agent is configured. ### Where can I find the AlohaJet CLI and MCP source code? The public [AlohaJet CLI repository on GitHub](https://github.com/AlohaBrowser/alohajet-cli) is licensed under Apache-2.0\. It includes command-line and MCP tools for browser control. This repository is specific to the CLI/MCP project and does not encompass the full desktop browser application. The README provides details on supported modes and limitations.