9.5. Research & Summarization¶

Researching a topic often involves juggling multiple browser tabs, copying text between apps, and manually organizing notes. Backend.AI GO's Cowork menu turns this into a single, streamlined workflow: describe what you want to research, and the agent autonomously searches the web, fetches articles, analyzes data, and produces a structured report—all on your desktop.

This guide walks through setting up Cowork for research tasks and provides a step-by-step example of generating a research report.

Why Use Cowork for Research?¶

Prerequisites¶

Before you begin, make sure you have:

• Backend.AI GO installed and running
• At least one model loaded — a capable model (7B+ parameters recommended) for best summarization quality
• (Optional) A Brave Search API key for web search — see Setting Up Web Search below

Built-in Tools for Research¶

The Cowork menu provides several built-in tools that are particularly useful for research workflows. No plugins or extensions are required.

Setting Up Web Search¶

The web_search tool uses the Brave Search API, which requires an API key.

1. Go to brave.com/search/api and sign up for a free API key (the free tier provides 2,000 queries per month).

2. In Backend.AI GO, open Settings > Tools & Extensions.

3. Enter your Brave Search API key in the designated field.

Step-by-Step Example: Technology Research Report¶

This example demonstrates a complete research workflow using the Cowork menu.

Step 1: Open Cowork¶

Click the Cowork icon in the sidebar to open the Cowork interface.

Step 2: Set Up Folder Permissions¶

The agent needs a place to save its research output.

1. Click the Folder Permissions toggle in the task input area.

2. Click Add Folder and select your research output folder (e.g., ~/Documents/research).

3. Set the permission level to Read & Write so the agent can save files.

Step 3: Configure Global Instructions (Optional)¶

Setting global instructions helps the agent follow a consistent research style.

1. Open the Settings drawer (gear icon in the header).

2. Go to the Instructions tab.

3. Enter instructions such as:

   Write in professional, clear prose.
   Always include source URLs for claims.
   Organize findings with headers and bullet points.
   When summarizing articles, include the publication date.
   

4. Enable the instructions and close the drawer.

These instructions will apply to all future Cowork tasks until you change them.

Step 4: Enter the Research Task¶

In the task input at the bottom of the Cowork page, describe your research goal:

Research the current state of WebAssembly adoption. Search for recent articles and technical blog posts, fetch the top 3 most relevant pages, and create a structured summary report saved as wasm-research.md in my research folder. Include key statistics, major use cases, browser support status, and notable projects using WebAssembly in production.

Press Enter (or click Start Task) to begin.

Step 5: Watch the Agent Work¶

The agent breaks the task into steps and executes them autonomously. You can watch the progress in the step viewer:

1. Web Search — The agent uses web_search to find recent articles about WebAssembly adoption.

2. Fetch Articles — It selects the most relevant results and uses fetch_url to retrieve the full content, which is automatically converted to Markdown.

3. Analysis — The agent reads through the fetched content, identifies key themes, statistics, and notable projects.

4. Report Generation — It synthesizes the findings into a structured Markdown report.

5. Save — The agent uses write_file to save the report to your research folder.

6. Summary — A final summary is presented with the key findings.

Step 6: Steer the Agent Mid-Task¶

If the agent is heading in a direction you want to adjust, use the Steering input that appears while the agent is running:

• "Focus more on edge computing and serverless use cases"
• "Also look for performance benchmarks comparing WASM to native code"
• "Skip the browser support section, I already know about that"

The agent incorporates your guidance without restarting from scratch.

Example: Summarizing Local Documents¶

Cowork is also effective for summarizing documents you already have on disk—no web search needed.

1. Add your documents folder with Read Only permission (the agent only needs to read).

2. Add an output folder with Read & Write permission for the summary.

3. Enter a task like:

   Read all the PDF files in my documents folder, summarize each one in 3-5 bullet points, and create a combined summary document saved as document-summaries.md in my output folder.

The agent will use read_file to read each document and write_file to save the combined summary.

Example: Analyzing Research Data¶

For data-driven research, Cowork can combine data analysis tools with summarization.

1. Add your data folder with Read Only permission.

2. Add an output folder with Read & Write permission.

3. Enter a task like:

   Analyze the survey results in survey-results.csv. Break down responses by demographic group, calculate key statistics (mean, median, standard deviation), and write a research memo saved as survey-analysis.md summarizing the findings.

The agent will use csv_reader to parse the CSV, run_python for statistical calculations, and write_file to save the analysis.

Advanced: Multi-Source Research¶

For comprehensive research, you can ask the agent to combine multiple source types:

Research renewable energy trends. Search for 3 recent articles online, read the local file energy-data-2024.csv for historical data, and create a report that combines online findings with data analysis. Save it as energy-report.md.

The agent will:

1. Use web_search and fetch_url for online sources
2. Use csv_reader or read_file for local data
3. Use run_python if calculations are needed
4. Synthesize everything into a unified report

Tips¶

• Start with a clear, specific prompt. The more detail you provide about what you want (format, depth, specific topics), the better the results.
• Use folder-specific instructions to encode project-level standards. For example, attach instructions like "Use APA citation format" to your academic research folder.
• Grant Read Only access to reference materials. If you have existing research or data files the agent should consult but not modify, add them as a separate folder with Read Only permission.
• Leverage Steering for iterative research. You don't need to get the prompt perfect on the first try—guide the agent as it works.
• Review and verify. AI-generated research is a powerful starting point, but always verify key claims and statistics against the original sources.

Troubleshooting¶

Related Pages¶

• Cowork Overview — How the Cowork agent works (ReAct reasoning, autonomous execution)
• Tools & Permissions — Full reference for built-in tools and permission settings
• Running Models — Load and manage local models
• Cloud Integration — Connect cloud models for enhanced capabilities