Discovery Agents are always-on scouts. You tell one what to watch, and it scans the literature on a schedule, ranks what it finds against your field and your library, writes a one-line reason each finding matters, and keeps it all in one place. They live under Radar → Discovery agents, and their findings are also a Discovery panel inside the editor, so you can add a result to your library, or add it and drop a citation, without leaving your document.
Discovery Agents are part of Fellow (3 agents) and Chair (10 agents). Every account on those plans gets an automatic Matched to your library agent, seeded from your research field (or, if you haven't set one, from your library) and ranking what it finds by how well it fits your library.
Create an agent
Open Radar → Discovery agents and choose New agent. Tell it what to track: a specific topic works far better than a broad one ("retrieval-augmented generation for clinical notes" beats "AI in medicine"). Then pick a persona. Personas aren't just for looks: each one carries a hunting style that pre-fills the agent's settings (how far back it looks, how many citations it requires, how often it runs, and how deep it searches). You can override any of it.
What each finding shows
For every result, the agent gives you:
- A one-line reason it matters: the specific contribution, written for your field.
- The abstract, as a short preview.
- Related to your library: a badge when a finding connects to something you've already saved.
- Actions: Add to Library, Read later (adds it to your library and your reading queue), Check vs work (does it support or challenge your drafts?), and Dismiss. Dismissing teaches the agent a little: after you pass on several results from the same venue, similar results from it rank lower in later runs.
You can sort findings by relevance, newest, most cited, or recently found, and filter to the new ones, open-access ones, ones already in your library, or ones related to it. Select two or more findings and choose Synthesize matrix for a comparison table of their methods, samples, findings and limitations.
The agent console
Each agent has a console with three tabs:
- Briefing. A short narrative of the themes across recent findings, refreshed on the schedule you choose: daily for fast-moving topics, weekly otherwise. Choose Email it to me to have it sent to you, and on Chair you can include narrated audio to listen on the go. Past narrations are saved so you can replay or share them.
- Ask. Chat with the agent about its findings: "summarize the themes," "which of these challenge my thesis?" Conversations are saved, so you can pick up where you left off without re-running anything.
- Trends. A timeline of findings by publication year and the recurring topics across them, with a small "rising" flag for topics concentrated in the last two years.
Deep scan (Chair)
By default an agent searches scholarly databases. Set the scan depth to Deep and it also searches the live web with an AI agent, surfacing preprints, lab pages, and reputable write-ups that the databases miss. Deep findings are tagged so you know where they came from.
Across your agents
If you run more than one agent, the agents page shows an aggregate view of trends across all of them, plus Synthesize across agents: one summary of the common threads connecting your different topics.
Share or follow an agent
- Share a public source list publishes the papers an agent found (titles, authors, journals, links) on a public page. Your briefing and the reasons stay private. It needs your public profile to be on.
- Turn on the RSS feed in an agent's settings to get a private link you can drop into any reader, so an agent's findings show up alongside everything else you follow.
Settings
Open any agent's settings from the gear on its card, or the Settings button on its page. You can rename it, change what it tracks, adjust recency and citation thresholds, switch personas, set the run schedule and the findings-email cadence (and send yourself a test email), post briefings to a Slack webhook, manage the RSS feed, and delete it. The automatic Matched to your library agent can be tuned but not deleted.
Each agent's email settings live with the agent. Account settings → Notifications links to them, and its Pause all non-essential email switch pauses agent emails along with the rest.