Skip to main content
Deep research mode goes beyond a single web search. NIOM fetches multiple sources, reads full pages, cross-references information, and synthesizes a comprehensive report — with citations for every claim.

​
How to trigger it

Just describe what you want researched. The Skill Tree routes to deep research automatically:
Say this…What happens
”Research quantum computing deeply”One-shot deep research → report in the conversation
”Research AI trends every 2 days”Creates a scheduled task that runs deep research on each execution
”Give me a deep dive on competitor pricing”Deep research with business-domain context
You don’t need to say “use the deep research tool.” The Skill Tree detects research intent and selects the right tools.

​
How it works

Your research request
  → Web search (multiple queries)
    → Fetch top sources (full page content)
      → Cross-reference and synthesize
        → Generate report with citations
          → Save as artifact (if in a task)
1

Search phase

NIOM generates multiple search queries from your request and runs them through the search provider chain (DuckDuckGo by default, upgradeable to Tavily or Brave).
2

Fetch phase

Top results are fetched and their full content is extracted using Mozilla Readability — the same engine Firefox uses for Reader View. Raw HTML is converted to clean, readable text.
3

Synthesis phase

The AI reads all fetched content, cross-references claims across sources, and produces a structured report. Key findings are attributed to their sources with URLs.
4

Output

The final report includes:
  • Summary — key findings in a concise format
  • Detailed analysis — organized by topic or question
  • Sources — every claim linked to the page it came from
  • Artifact — if running in a task, the report is saved as a .md file you can retrieve later

​
Deep research in background tasks

The real power is combining deep research with background tasks: 
"Research the latest AI breakthroughs every week and save a summary"
This creates a recurring task that:
  1. Runs deepResearch on a schedule
  2. Saves each report as an artifact
  3. Streams progress live via SSE — you see sources being fetched in real-time
  4. Can be steered: “Focus more on open-source models next time”
Each run builds on the task’s memory, so reports improve over time.

​
Search provider chain

Deep research uses the same search provider chain as regular web search:
PriorityProviderAPI key needed?Quality
1Tavily (if configured)Yes (free: 1,000/month)Best — purpose-built for AI research
2Brave (if configured)Yes (free: 2,000/month)Great — fast, rich snippets
3DuckDuckGo (always available)NoGood — works out of the box
If your preferred provider fails, NIOM falls back automatically. You always get results.

​
Viewing results

​
In a conversation

The deep research output renders with rich formatting:
  • Topic and depth level
  • Source count and quality indicators
  • Individual sources with titles, URLs, and relevance

​
As an artifact

If the research was done in a task, the full report is saved as a Markdown file. Find it in the task’s artifact panel or at ~/.niom/artifacts/.
Deep research works best with capable models like Claude Sonnet or GPT-4o. Fast models (Groq, etc.) may produce shorter, less thorough reports.