Research the live web, with citations The /research command runs Krater's deep research agent. It plans sub questions around your topic, searches the web for each one, reads the most relevant pages...
The /research command runs Krater’s deep-research agent. It plans sub-questions around your topic, searches the web for each one, reads the most relevant pages, and hands you a written report with numbered, clickable sources — all inside your chat.
For quick lookups you do not need a command at all: when a normal chat question needs fresh information, the agent searches the web automatically for about 2 credits and answers with citations. Use /research when you want a real investigation, not a single search.
Cost: 13 credits standard depth (the estimate is upfront — any unused amount is refunded). A thorough depth is available for 21 credits. Research requires a paid plan (Pro, Ultra or Max) or a team seat.
Type /research followed by your question:
/research What are the latest developments in quantum computing?
/research Compare pricing and features of the top 5 AI coding assistants
Ask for “thorough” or “deep” research in your prompt when the topic needs more coverage:
/research Run a thorough investigation of the GPU cloud market in 2026
Most runs finish in about two minutes.
You can expand the “Worked for Xs” log during the run to see each search and read. Afterward, the Sources button on the answer opens the full citation list.
| Standard | Thorough | |
|---|---|---|
| Search calls | 5 | 8 |
| Pages read | Up to 6 | Up to 10 |
| Cost | 13 credits | 21 credits |
| Best for | Most questions | Market maps, literature reviews, high-stakes checks |
Costs are the upfront estimate; any unused reservation is refunded automatically.
| Quick web search (automatic) | /research |
|
|---|---|---|
| How it runs | The agent decides your question needs fresh info | You type /research |
| Searches | 1 | 5–8, planned |
| Pages read | Snippets only | Up to 10 full pages |
| Cost | ~2 credits | 13–21 credits |
| Output | Short cited answer | Full cited report |
Research:
Fact-checking:
Discovery:
Learning:
| Tip | Example |
|---|---|
| Be specific | “React 19 new features” vs “React” |
| Add context | “best laptop for video editing under $1500” |
| Use timeframes | “AI funding rounds Q3 2026” |
| State the output | “…as a comparison table with sources” |
After a report, you can keep asking in the same chat — the agent remembers the research context: