Research is a web app for deep online research, planning and documentation. Give it a topic and it runs dozens of searches, downloads and reads hundreds of pages, pulls out structured data and assembles a verified report into a document library. Agent works in the background at the same time — its chats are visible and can be managed from the same interface.
Who it’s for
Research is built for market analysis, planning and documentation — no development skills needed. Unlike Agent, which looks and works like an IDE, everything here speaks the language of research: topics, search sessions, downloaded pages, reports and plans.
System requirements
The product runs on a Linux server and is deployed through MultiAgent, as part of the Agent Factory subscription: Docker and everything else is installed by MultiAgent itself. Data lives in the database built into the product, or in PostgreSQL — your choice.
First sign-in
The first user registers through the web form on the login page: login, password (4 characters or more), name and email. The password is stored as a SHA-256 hash, and the session runs on JWT tokens. For automatic deployments, the same user can be set in advance with the RESEARCH_APP_LOGIN and RESEARCH_APP_PASSWORD environment variables — the password there is set as a SHA-256 hash, and you can compute it right on the login page.
Home screen
- Documents — your research library: a tree of Markdown documents organized by category;
- Searches — search sessions and their results;
- Fetches — download sessions and the Markdown they fetched;
- Completions — the log of LLM calls, with token usage and cost;
- Sessions — research sessions: the searches, downloads and extractions behind one topic;
- Chats — Agent chats updating in real time; opened in a panel by the button at the bottom of the sidebar;
- Services — search providers, download methods and LLM services;
- Scripts — the product extends with your own Python scripts, no rebuilds needed: this screen shows the environment state, the definitions, and the run journal (details in “Python Scripts“).
Your first research
- set up a search provider (its API key) and an LLM service in the Services section;
- the library starts from scratch: on an empty tree the first research creates the first category by itself, and from then on reports land in it;
- give the agent a research topic — it will run the searches, download the pages and analyze them;
- watch the progress: every operation shows a status and an estimated completion time;
- read the finished report in the Documents library — together with the search sessions and the source pages.
It pays to prepare several topics in advance: put them into the prompt queue — the agent will work through the tasks one after another on its own (see “Prompt Queue“). A long task doesn’t have to be typed either — you can dictate it (“Voice Input“).
Interface
The interface language switches between English and Russian in the settings, and the theme between dark and light (light is the default). Research events come with sound cues — you can hear how the work is going even when the tab is in the background. A sound is played once, even when the app sits open in several tabs or windows: there are no repeats, and a badly delayed sound is not played at all.
What’s next
How deep research works is covered in the Online Research Skill section. Searching and downloading are covered in Search and Fetching Pages, and data extraction in Data Extraction. Working with documents is described in the Research Library.