Research Agent

The built-in Research Agent researches the web and delivers a structured answer with sources.

How it works

1. Select the Research mode in the composer.
2. Enter your question and send it.
3. The agent plans a research strategy, searches and reads web pages, and summarizes the findings.

Run trace

You can follow the research steps live in a dedicated view: the plan, web searches, sources read, and the final answer. At the end, the credits consumed are shown.

Built-in web search (MCP)

The Research Agent uses a built-in web search via the open Model Context Protocol (MCP). Our own research server sends the search queries to the Brave Search API and returns real search results in real time. What is transmitted is described in the privacy policy.

• One-time consent: Before the first use, a consent dialog asks whether external tools may be used. The decision applies per service and can be changed at any time.
• Transparent and controlled: Every tool call (search, page fetch) appears as a step in the run trace. Only search queries and URLs are sent, never file contents or credentials.
• Safe fallback: If the research server is unavailable, web search may fall back to the built-in search of our AI provider instead of the agent failing.

Research documents

• After a successful run, a research document is saved automatically.
• Documents appear in the sidebar (per project, or under "Recent").
• They are reusable as project knowledge and can be attached as chat context.
• Sources are cited inline ([1], [2], ...) and listed as a numbered source list at the end.
• Every completed agent result can be sent by email (button in the result view; recipient prefilled with your own address).

Cost

• Affordable research with GPT-6 Luna.
• Premium research with Claude/GPT (considerably higher).

Billing uses credit hold/settle; only actual usage is charged (capped at the hold). Research requires an active subscription.