A general AI answers from memory. Real research still means a dozen open tabs, and no sources to stand on.
Ask the live web anything, get one grounded answer.
A manager agent routes your question to specialist scouts (web, social, developer, and media), then reconciles what they find into a single sourced answer. One team in place of a dozen open tabs.
"What is Acme Robotics shipping and saying this quarter?" Route to Web + Social + Dev scouts.
Series B closed in March, new ops hires, pricing page updated.
CEO posting about warehouse automation launch on LinkedIn.
Acme is scaling ops post-raise, launching warehouse automation. 4 sources attached.
one run · manager plans, specialists report, you get one answer
The job, before you had a team for it.
This is the work Research Desk quietly takes off your plate.
The answer lives across a dozen tabs and three logins.
By the time you've stitched it together, half of it is stale.
One generalist prompt guesses instead of actually looking.
Nobody can tell which claim came from where.
A team of agents, not one prompt.
A manager agent reads the goal and hands each part to a specialist built for it, then reconciles their work into one result.
Reads your question, decides which scouts to call, dispatches them, and merges their findings into one answer with sources.
Searches the open web and reads any URL to answer questions that don't belong to a single platform.
Pulls what people and companies are saying right now across X, LinkedIn, and Instagram.
Tracks trending repositories, open-source projects, and research papers across GitHub, Hacker News, and arXiv.
Fetches live data on videos, channels, and topics from YouTube.
How it runs.
Fork it, point it at your inputs, and let the team work.
Type a question or wire one in from an earlier step. No query syntax, no operators.
It decides which scouts are relevant and dispatches them, in parallel where it can.
Each specialist queries its own sources and returns grounded, current findings.
The manager reconciles everything into a single sourced response, ready for the next step.
Ask about a prospect before an outreach.
"What is Acme Robotics working on right now, and who runs ops?"
- Series B raised in March; hiring across operations.
- Just announced a warehouse-automation product line.
- Head of Ops: named, with a link to the source.
- Four sources attached, each mapped to a claim.
Where teams put it to work.
Brief a rep on a company minutes before the call, with sources they can trust.
Track what competitors are shipping and saying without living in a feed reader.
Drop it mid-run so a draft or a decision is built on current facts, not stale memory.
- A question in plain language
- Optional focus: a company, topic, or platform
- A single grounded answer
- The sources behind every claim
- Platform-specific findings, kept separate when you need them
- A summary clean enough to drop into the next step
What is an AI research assistant?
An AI research assistant is a tool that reads the live web on your behalf and returns a sourced answer, instead of just a page of links. You ask a question in plain language, and it does the searching, reading, and reconciling for you.
The best ones do not answer from memory. They search current sources at run time, keep the citations, and reconcile conflicting results into one clear answer. Research Desk goes a step further and splits the work across specialist agents for the web, social, developer, and video sources.
How is an AI research assistant different from AI search?
AI search runs one query and returns one answer. An AI research assistant like Research Desk runs a team: a manager routes the question to specialist scouts, each digs into its own sources, and the manager reconciles everything into a single sourced report.
That is the difference between a quick lookup and real research. For a fast fact, a single search tool is enough. For a question that spans platforms, or that feeds a decision, the multi-agent approach gives you breadth and a resolved answer, not one thread you still have to verify.
New to this? Read our guide on how to build a sales battlecard.
Questions, answered.
How is this different from one AI with web search?
A single agent runs one search and hopes. Research Desk routes to specialists tuned for the web, social, developer, and video sources, then a manager reconciles them, so you get breadth and a resolved answer, not one thread.
Can I trust the answer?
Every claim comes back with the source it was built from. When the live web doesn't support an answer, the team says so instead of guessing.
Can I use it inside a workflow?
Yes. Fork it and drop it in as a step, so any run can pause to ask the live web before it acts.
Fork Research Desk and make it yours.
Start from a working team of agents, then swap the tools, tune the instructions, and run it in your own workflows.
Fork this template