Use Deep Research Modes
Deep research can be tuned based on how thorough you need the output to be. The pipeline always runs the same four phases (SCOPE → RESEARCH → EVALUATE → SYNTHESIS), but the scope and depth vary.
Standard mode (default)
The default mode decomposes a topic into 3–6 dimensions and runs 1–2 evaluation iterations.
- When to use: Most research questions — technical topics, comparisons, surveys
- Time: 2–5 minutes depending on topic breadth
- Sources per dimension: Up to 8 (primary web + secondary social/IT search)
Quick mode
Restricts to 2–3 dimensions with no gap-filling iteration.
- When to use: You need a fast overview, or the topic is narrow
- Time: Under 2 minutes
- Sources per dimension: Up to 5 (primary web only)
To request a focused result, phrase your query narrowly:
quick overview of WebAssembly in the browserComprehensive mode
Generates 5–8 dimensions and runs up to 3 evaluation iterations. Secondary searches use multiple category combinations.
- When to use: Deep dives, literature surveys, architecture decisions
- Time: 8–15 minutes
- Sources per dimension: Up to 12 (primary + secondary + pagination)
To trigger deeper research, phrase your query broadly with explicit scope:
comprehensive analysis of distributed tracing approaches for microservices at scaleControlling iteration depth
The gap-filling loop runs up to 3 times. If the LLM evaluator finds gaps after each round, it spawns new sub-dimensions.
You can see how many iterations ran in the activity feed:
Evaluation: 1 gap found. Starting iteration 2...
Evaluation: no gaps. Proceeding to synthesis.Cancelling a running job
Click Cancel in the research UI, or send:
DELETE /api/research/deep/:idThe job stops cleanly after the current dimension completes. Already-gathered sources are preserved.
Downloading results
Click Download (top right) to export as Markdown. The export includes:
- Full report text with inline citation markers
- Numbered source list at the bottom
- Research metadata (topic, duration, source count)
Related
- Deep Research API — programmatic control of research jobs
- Deep research pipeline — how each phase works internally