Research Agent

Three modes. Set the depth. Walk away.

Every research session starts with a choice of depth. The agent reads the brief, selects sources, and begins writing. The report builds in real time on the right panel — you can watch it take shape or come back when it is done.

Quick

5–10 sources

Fast, focused answers. Good for confirming a fact, getting oriented on an unfamiliar topic, or when you need something in under a minute.

Deep

20–50 sources

Comprehensive investigation. The agent searches multiple angles, cross-references sources, and builds a structured report with key findings, analysis, and a reference list.

Marathon

Unlimited

Extended research that can run for hours. Save mid-session and continue from exactly where the agent stopped — no restart, no lost context.

The agent reads broadly. Keeps only what matters.

The Research Progress panel tracks two numbers that tell the whole story: raw tokens collected vs. active context in use. The agent pulls in everything it can find, then compresses aggressively — keeping only the information that advances the research brief.

Raw Collected

8,940

tokens gathered across all sources

Active Context

306

tokens actually in working memory

Compression Ratio

29:1

signal kept, noise discarded

This is why the Research Agent can cover 50 sources without burning through your context window. It is not retrieving whole pages — it is extracting the fragments that matter and stitching them into a coherent document.

Watch it work. Every step, visible.

The live session panel on the left shows every tool call as it happens: each search query issued, each URL fetched, each note added to the draft. The collapsible tool-call list expands to show the full detail. The confidence percentage, source count, and search status update in real time.

Searching Agent issues targeted search queries across multiple engines. Every query is logged and visible in the tool-call list.

Fetching & Extracting Each promising URL is fetched. The agent extracts relevant content, discards noise, and adds the useful material to its working context.

Writing The live report builds in the right panel as the agent works. Sections appear, tables render, bullet points populate. The document is taking shape before the research is even finished.

Complete Final report with key findings, structured analysis, open questions, and a full reference list. Export as markdown or copy directly to the clipboard.

Sources tracked.

Every source the agent reads is logged in the Sources panel: URL, page title, domain, the time it was fetched, and a quality rating. Nothing is cited without being traceable.

Save. Continue. Come back to it.

Marathon research sessions can run for hours. Once a session finishes you can save the report, then continue it later — the agent picks up from where the first run ended and keeps going, extending coverage of the same topic without starting over.

Continue from saved — reload a session and keep going
Export — download as markdown or copy to clipboard
Reports tab lists every past session with filter, view, and delete

Document Intel — search what you've already researched.

The Data tab is a workspace-wide document search engine. Every report you've saved, every document in your workspace — the agent can search it semantically. Results are grouped by report, then document, then the specific chunk that matched.

Click a result and it opens the original document in the editor at the exact character position where the match occurred. No manual scrolling, no copy-pasting. If the source is a PDF it opens in the PDF viewer.

Persistent Search History