> ## Documentation Index > Fetch the complete documentation index at: https://docs.blink.so/docs/llms.txt > Use this file to discover all available pages before exploring further. # Research Agent > Build a Blink agent that acts as your research and web browsing partner. ## Overview Blink lets you talk to Edit Mode and describe the agent you wish you had. For research and web browsing work, the goal is to craft a partner that can scan the web, compare findings, keep track of promising leads, and report back in the style you prefer. This page focuses on the creative side: what capabilities to request, what personality to give your agent, and concrete prompts you can copy and paste to get started quickly. ## Getting started 1. Create a new agent directory and run `blink init` 2. Start the dev server with `blink dev` 3. Press `Ctrl+T` to toggle to Edit Mode 4. Describe your research assistant using the ideas below ## What you can build Here are examples of what you can ask your research agent to do once you've built it. These show the end result: scroll down to see how to build an agent that can handle these requests. ``` Do a landscape scan of the top 3 React state management libraries actively maintained in 2024. For each, include: GitHub stars, last update date, key differentiator, ideal use case, and major drawbacks. Provide links to official docs. ``` ``` Compare Docusaurus vs Mintlify for developer documentation. I'm a startup with 2 engineers. Focus on: setup time, customization options, cost, and ongoing maintenance burden. Make a recommendation. ``` ``` I read that Next.js 14 improved performance by 40% with server actions. Can you verify this claim? Find the original source and check if other benchmarks confirm it. ``` ``` Explain how Vercel's edge runtime works. ``` Then follow up with: ``` Can you find some example repos that use edge functions effectively? ``` And: ``` What are the main limitations I should know about? ``` ``` Check all the links you just shared. Let me know if any are dead, paywalled, or redirecting unexpectedly. ``` ``` Summarize everything we've discussed about database options for real-time apps. Format it as a brief I can share with my team on Slack. ``` ## Building your research agent Switch to Edit Mode (`Ctrl+T`) and use these prompts to build your agent. Copy and customize them based on what you want your agent to do. You don't need to use all of these prompts. Start with 2-3 that match your needs, test the agent, then add more capabilities as needed. ### Make it search and gather information ``` Add a web search tool so the agent can find current information online. It should search for 5-10 results and extract the full text content, not just titles and snippets. ``` ### Make it cite sources properly ``` Every claim the agent makes should include an inline citation like [1] [2], with full source details at the end. If I ask for elaboration on a specific point, it should cite additional sources. ``` ### Make it compare and analyze ``` Give the agent the ability to compare multiple sources on the same topic. It should create a comparison table showing: key claims, level of detail, recency, and whether sources agree or contradict each other. ``` ### Make it collaborative and friendly ``` Give the agent a friendly, collaborative tone. It should act like a helpful research partner, not a know-it-all. Before diving into research, it should summarize its research plan and ask if the approach sounds right. ``` ### Make it verify quality ``` Add guidance so the agent double-checks any link before sharing it. It should warn me if a source looks outdated (more than 2 years old), paywalled, or from a questionable domain. For technical topics, it should prefer official documentation and established tech blogs. ``` ### Make it ask clarifying questions ``` The agent should ask clarifying questions before starting broad searches. It should confirm: target audience, timeframe (current info vs historical), desired level of technical depth, and whether I need sources I can share publicly. ``` ### Make it track research over time ``` Add a notes system that persists across messages. For each research topic, track: key findings, unanswered questions, best sources found so far, and recommended next actions. Let me ask "show me notes on [topic]" at any time. ``` ### Make it format reports consistently ``` When presenting findings, the agent should use this structure: 1) Executive summary (2-3 sentences), 2) Key findings as bullets, 3) Full sources list with titles and URLs, 4) Suggested next steps or open questions. ``` ### Make it show confidence levels ``` The agent should indicate confidence levels for findings: "High confidence" (3+ agreeing sources), "Medium confidence" (1-2 sources), or "Uncertain" (conflicting info or single questionable source). Always explain why. ``` ### Add helpful slash commands ``` Add slash commands: /summary for quick overview of current research, /sources to list all references, /questions to see what's still unclear, /export to format everything as a shareable document. ``` ## How it works behind the scenes When you use the prompts above, Edit Mode automatically implements the right tools for your agent. You don't need to code anything; just describe what you want. Common tools include: * **Web search** - Find current information across the internet * **Content extraction** - Pull full text from URLs and PDFs * **File operations** - Read documents, save research summaries and reports * **Bash execution** - Run scripts to download or process data Edit Mode handles all the technical implementation. Your job is to describe the agent's behavior and capabilities. ## Iterating on your agent As you test, you'll discover what works and what needs adjustment. Use these Edit Mode prompts to refine your agent's behavior: ``` The responses are too long. Keep summaries under 200 words and use bullet points for details. Only include the top 3 most relevant sources unless I ask for more. ``` ``` The agent isn't consistently citing sources. Make it mandatory: never make a factual claim without a citation. If no source exists, say "I couldn't verify this." ``` ``` Before searching, the agent should ask me to narrow the scope. For example: "Should I focus on solutions compatible with TypeScript?" or "Do you want enterprise or open-source options?" ``` ``` Prioritize these sources: official documentation, established tech blogs (like Vercel blog, GitHub blog), recent conference talks, and papers from known institutions. Avoid: random Medium posts, outdated Stack Overflow answers, marketing sites. ``` ``` For technical topics, the agent should read full documentation pages, not just skim landing pages. It should look for: architecture decisions, performance characteristics, known issues, and community sentiment. ``` ## Advanced capabilities Once your basic agent works well, consider these enhancements: ``` Create a research workflow: 1) Initial broad search to identify top options, 2) Deep dive on the top 3 with pros/cons, 3) Fact-check key claims, 4) Generate comparison table, 5) Make recommendation with rationale. Let me approve or adjust the direction after each step. ``` ``` Implement a credibility scoring system. Official docs = 10 points, established tech blogs = 8, GitHub repos with 1k+ stars = 7, recent Stack Overflow answers = 5, random blogs = 3. Only cite sources scoring 5+. Show scores in citations. ``` ``` Add a monitoring mode: I should be able to say "track developments in [topic]" and the agent will periodically search for new information, then notify me of significant changes or announcements. ``` ``` Create templates for common research tasks: "framework evaluation", "vendor comparison", "security audit", "migration planning". Each template should define the structure, key questions to answer, and required source types. ``` ## Tips for success **When you're building:** * Start with 2-3 capabilities, not all of them * Give Edit Mode examples of the output you want to see * Make one change at a time, then test it * Save prompts that work well for future reference **When you're ready to scale:** * Deploy your agent via web or Slack for team access * Add persistent memory to track research across sessions * Connect to specialized APIs or internal databases * Share your agent configuration with teammates Focus on getting the basic behavior right first. You can always add more sophistication later.