{"id":17001,"date":"2025-04-01T08:00:17","date_gmt":"2025-04-01T15:00:17","guid":{"rendered":"https:\/\/www.couchbase.com\/blog\/?p=17001"},"modified":"2025-06-13T16:39:40","modified_gmt":"2025-06-13T23:39:40","slug":"build-your-first-open-source-ai-agent-with-couchbase","status":"publish","type":"post","link":"https:\/\/www.couchbase.com\/blog\/build-your-first-open-source-ai-agent-with-couchbase\/","title":{"rendered":"Build Your First Open Source AI Agent with Couchbase"},"content":{"rendered":"

If 2024 was the year of AI chatbots, then 2025 is the year of AI agents. At first glance, they may seem similar, but nothing could be farther from the truth. While you may interact with an AI agent in the same manner you interact with an AI chatbot, perhaps through a web interface, the differences between them are stark.<\/span><\/p>\n

AI agents can autonomously act to fulfill your request. That\u2019s right, I said it: autonomously<\/em>. AI acting on your behalf figuring out what to do and what its next steps should be without your intervention.<\/span><\/p>\n

Want to find a good restaurant for a date night and book a table? Let your AI agent do it for you, from the research to the booking.<\/span><\/p>\n

Need to track new customer engagements in your CRM and send customized and personalized email outreach? Let your AI agent do it for you, from the tracking to composing the email all the way to sending it.<\/span><\/p>\n

Essentially, agents can do just about anything for you online. You can even have multiple agents working in tandem. All they need is a <\/span>task<\/span><\/i> that you set them off with and they will go about completing that task.<\/span><\/p>\n

You may think that building or using agentic AI will cost you a fortune, and you wouldn\u2019t be wrong for thinking that. Companies, some of the largest in the space, are offering these services for very high monthly fees. Yet, with some knowledge of Python and JavaScript, you can build a full stack AI agent web user interface for free. Best of all, you can easily incorporate Couchbase for performant data storage, retrieval and full text search capabilities. Couchbase Capella, the fully managed database as a service, even offers a fully free forever tier, that you can use in your application.<\/span><\/p>\n

\"\"<\/a>Did I pique your interest? Well, great news! You don\u2019t even have to build it from scratch. You can find the full application in my open source browser agent GitHub repository<\/a> – <\/span>with instructions on how to start using it.\u00a0<\/span><\/p>\n

Let\u2019s take a look at the architecture of the application, and how it is built. A better understanding of how it works will empower you to adapt it for your own personal use case.<\/span><\/p>\n

The application is made up of a React frontend and a Python backend. We\u2019ll focus on the backend as that is where the work of the agent and the interactions with Couchbase happen. You can explore the frontend freely on GitHub, which provides a mobile-first design.<\/span><\/p>\n

Exploring how it works<\/span><\/h2>\n

The foundation of this application is a harmonious blend of a React frontend and a Python backend. While the frontend ensures a smooth and responsive user experience, the backend is where the AI agent operates, handling user requests and interacting with Couchbase for data management.<\/span><\/p>\n

The AI agent with browser use<\/span><\/h3>\n

At the core of the AI agent is the integration of the open-source Python library from Browser Use<\/a>. This library empowers the agent to perform a variety of tasks by simulating browser actions. Whether it’s navigating to a restaurant’s website to book a table or accessing the CRM to track customer engagements, Browser Use provides the necessary tools for the agent to interact with web resources effectively.<\/span><\/p>\n

Initializing a Browser Use AI agent only takes a few lines of code in your Python backend:<\/span><\/p>\n

async def run_browser_agent(task: str):\r\n\u00a0\u00a0\u00a0\u00a0\"\"\"\r\n\u00a0\u00a0\u00a0\u00a0Runs the browser agent and returns its results in a JSON-serializable format.\r\n\u00a0\u00a0\u00a0\u00a0\"\"\"\r\n\u00a0\u00a0\u00a0\u00a0agent = Agent(task=task, llm=llm, use_vision=True, max_failures=3, retry_delay=5)\r\n\u00a0\u00a0\u00a0\u00a0history = await agent.run(max_steps=50)\r\n\r\n\u00a0\u00a0\u00a0\u00a0serialized_results = []\r\n\u00a0\u00a0\u00a0\u00a0if history and history.history:\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0for step in history.history:\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0if step.result:\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0if isinstance(step.result, list):\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0for res in step.result:\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0serialized_results.append(vars(res))\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0else:\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0serialized_results.append(vars)\r\n\r\n\u00a0\u00a0\u00a0\u00a0return serialized_results<\/pre>\n
The Agent is instantiated with several parameters, including:<\/span><\/p>\n