Documentation Index
Fetch the complete documentation index at: https://docs.snorbe.deskrex.ai/llms.txt
Use this file to discover all available pages before exploring further.
Autonomous Research Agent
The Autonomous Research Agent is an AI assistant that conducts research on your behalf.Available AI Models
You can change the AI model used by the agent from the model selector in the chat interface. You can pick a single model, but we recommend Snorbe Fast / Snorbe Quality — they automatically route each task to the best-fit model for you.Snorbe Fast / Snorbe Quality (recommended, auto-routing)
| Mode | Best for |
|---|---|
| Snorbe Fast | Trying things out, running large batch research, optimizing for speed and cost |
| Snorbe Quality | Maximizing output quality for investment decisions, research reports, and critical deliverables |
Per-task routing examples
| Step | Snorbe Fast | Snorbe Quality |
|---|---|---|
| Plan creation / final answer / plan decision / plan review | GPT-5.4-mini | GPT-5.4 |
| Chat routing / browser operation instructions | GPT-5.4-mini | GPT-5.4 |
| Graph extraction (nodes and relations) | GPT-5.4-mini | GPT-5.4-low |
| Keyword extraction / entity description merging | GPT-5.4-nano | GPT-5.4-mini |
| Image summarization | GPT-5.4-nano | GPT-5.4-mini |
| Long-context compaction summary | Gemini 3 Flash Preview | Gemini 3.1 Pro Preview |
Steps where reasoning matters (plan creation, final answer) use the stronger model in both modes. Lightweight extraction, merge, and summarization steps route to smaller models. Think of Snorbe Fast vs. Quality as “one tier up per step” rather than two entirely separate model stacks.
Individual model selection
If you want to pin the agent to a single provider model instead of using auto-routing, the following are available:| Provider | Model | Description |
|---|---|---|
| Anthropic | Claude 4.6 Sonnet | Balanced. Fast and accurate for everyday research |
| Anthropic | Claude 4.6 Opus | High precision. Strong for complex research and reasoning. Supports Adaptive Thinking |
| Anthropic | Claude 4.7 Opus | Latest top-tier model. Supports 1M-token long context and Adaptive Thinking |
| Gemini 3.1 Pro | Handles ultra-long contexts effectively. Great for cross-file analysis | |
| OpenAI | GPT-5.4 | Versatile balanced model for a wide range of tasks |
| xAI | Grok 4 | Strong real-time information processing |
What the Agent Can Do
Web Search
Search the internet for the latest news, research papers, and technical information. Automatically complete cross-source searches that would be difficult to do manually.
Web Scraping
Automatically extract detailed information from pages found through search. Aggregate necessary data from multiple sources.
Knowledge Graph
Organize collected information by relevance and visualize it in graph format. See connections between information at a glance.
Report Generation
Output research results as structured reports. Downloadable in Markdown or PDF format.
How to Use
1. Enter a Topic
Enter the topic you want to research in the chat. For example:- “Next-generation battery technology trends”
- “Drone delivery regulations”
- “Sustainable materials market dynamics”
2. Agent Automatically Starts Research
After entering your topic, the agent automatically creates a research plan and begins web search and information collection.3. Review and Utilize Results
Once research is complete, a knowledge graph and report are generated. You can also ask follow-up questions or dive deeper into specific areas.Real-Time Progress
While the agent is running, you can monitor progress in real-time. All processing steps are displayed transparently, so you can understand how the AI is working.Search Status
Shows which keywords are being searched and how many results have been found.
Referenced Sources
Displays the sources of collected information in real-time. Verify the credibility of information.
Processing Steps
Shows the current stage (search, extraction, organization).
Stopping Research
While research is running, a stop button appears next to the input field. Click it when the direction feels off or when you want to switch to a different instruction. Any information collected so far remains in the Knowledge Graph.Configuring Browsing Behavior
From the settings icon in the chat interface, you can adjust the maximum number of steps the AI takes when browsing websites (1–100). Increase the value for in-depth investigation, or decrease it for a quick overview.Giving Instructions During Browsing
While the agent is automatically browsing websites, you can provide additional instructions or answer questions from the agent in real time. You can interact naturally — for example, “check a different page instead” or “you’ll need to log in here.”Verifying Accuracy
If a research result leaves you wondering “is this really accurate?”, you can request verification using the fact-check button. The agent will re-examine the response and show the supporting evidence.Sending Feedback
You can send feedback, bug reports, or feature requests anytime using the feedback button in the sidebar. It reaches the development team directly, right when you notice something.Available Tools
The agent automatically selects and uses the following tools based on the situation.Recall
Search past research results and knowledge graph. No need to repeat the same questions.
Web Search
Search for the latest information on the internet. Covers news, papers, and technical articles.
X Search
Collect the latest trends and real-time information from X (formerly Twitter).
Source Summary
Enter a URL to summarize the page content. Efficiently understand multiple sources.
Browsing
Actually operate websites to collect information. Supports login-required sites and dynamic pages.
Extract Related URLs
Automatically crawl related pages from a specified site. Analyzes sitemaps and link structures.
Academic DB Search
Directly retrieve specialized information from arXiv, PubMed, Semantic Scholar, e-Stat, and other academic and statistical databases.
When you enter complex instructions that span multiple actions (e.g., “Research the latest AI information on social media, then do further web searches and finally summarize in a report”), the agent automatically activates Plan Mode and coordinates multiple tools to conduct research step by step.
Comparison with Manual Research
Traditional Manual Research
- Switch between multiple search engines
- Open each page to review content
- Copy and organize information
- Record citations
- Manually create reports
- Takes hours to days
Agent-Based Research
- Simply enter a topic
- Automatically collects relevant information
- Automatically organizes information relationships
- Citations recorded automatically
- Structured reports generated automatically
- Significantly faster than manual research
Concrete Deliverables
Perspective Matrix
Organizes information in matrix format across multiple perspectives such as materials, methods, costs, regulations, and market applications. Export to CSV or Excel for internal sharing or proposal documents.
Structured Report
Automatically generates a report including research objectives, summary, detailed analysis, and citation list. Downloadable in Markdown or PDF format.
Common Use Cases
Technical Research
Quickly collect information essential for R&D, such as new material properties, latest research papers, and competitive technology comparisons.
Market Research
Organize information needed for new business planning, including market size, growth rates, key players, and regulatory trends.
Competitive Analysis
Research competitor products, technologies, and patent filings to understand your company’s position.
Regulatory & Policy Research
Track changes in regulatory environments, including laws by country, subsidy programs, and industry guidelines.
Trend Analysis
Early detection of latest trends from social media and news, and changes in consumer behavior.
Proposal Generation
Create draft proposals based on research results, including comparison tables by perspective and recommendations.
Try It Now
Simply enter one topic to experience how the agent conducts research. No complex configuration required.