An AI research group that brings back evidence, not just answers
The One's Research group splits a question into branches that gather evidence independently, each on its own task, then writes one report with its sources. The work lives in a matter, whose board keeps the notes, links, files and data found along the way. The report comes back as a delivery you accept or send back. It can run once, or again every week when a page or a source changes.
What you can hand over
- “Compare the five most-used open-source vector databases on licence, hosting and price, with sources.”
- “Find what changed in the EU's rules on AI-generated images this year, and what it means for an online shop.”
- “Every Monday, check these three competitors' pricing pages and tell me what changed.”
How the group works
- Independent branches. Researchers take separate questions from a task board, so one branch's first guess does not steer the others.
- On the web. Agents search the web through your model provider's search, or your own Tavily account with the Tavily Search Plugin, and read pages in a Chrome profile of their own.
- Your files count too. Point the work at a folder and the group reads what is in it.
- It asks you. When the question is ambiguous, the group asks; nobody answers for you.
What comes back
A report with its sources, checked against what the matter asked for. The matter's board keeps what the group collected: notes, links, files and data. Accept, Rework or Give up.
Research that repeats
Give the matter a cadence: every week, or when a web page changes (only the lines that match a pattern, if you like). Each fire goes to the same group, which keeps what it learned last time. See Recurring work.
Where it stops
- Pages behind your own sign-ins are not read through your browser's sessions: the group's Chrome profile is its own.
- Agents stop when your computer is off.
- Facts the group finds are as good as its sources; the report shows them so you can check.
Questions
- Is this the same as “deep research” in a chat app?
- It does the same kind of work, but on your computer, with your files and models, and it can keep a question open for weeks on one board.
- Can it use local models?
- Yes, through a compatible endpoint such as Ollama, LM Studio or vLLM. A local model has no web search of its own, so search falls back to a keyless engine; add a Tavily key for dependable search.
In the docs