Discover

What's new in the world for your product's domain.

Research, industry and trends sources — auto-configurable from your topics — producing an editable report with highlights, themes and categories. Streams step-by-step progress.

What it is

Your product's news feed, curated by agents.

Discover pulls from broad research, industry and trends sources (Hacker News, arXiv, company blogs, Product Hunt, Hugging Face, Google Trends, Exploding Topics, and more). Each item is passed through a cheap pre-filter that tags it with labels like trend, pain point, feature request, bug report, and writes a one-line summary. The agent then generates a module report: highlights, themes, and categories.

How it works

Three steps: configure, collect, generate.

Smart-configure reads your product's topics and activates matching sources. Collection runs on a schedule (or on demand). Generation streams back step events over SSE.

Smart-configure

POST /modules/discover/smart-configure

The agent reads the product's topics and competitors and proposes which sources to activate (e.g. specific subreddits, arXiv categories, company blogs). You approve before anything runs.

Collect

POST /modules/discover/collect (SSE)

Collectors run in parallel. Each item goes through a Haiku-based pre-filter that labels it and writes a short summary. Items below the relevance threshold are dropped.

Generate

POST /modules/discover/generate/stream (SSE)

The module report is produced and streamed event-by-event. Regenerate on demand; every prior report is kept in history so you can compare week-over-week.

Ask

POST /modules/discover/chat

The module has its own chat tab — ask ad-hoc questions against the filtered items without leaving the page.

SSE events

source_start — source N is collecting now
source_complete — source finished, returns item count
prefilter_progress — pre-filter pass progress per item
step — report generation step (highlights → themes → categories)
done — whole run finished

Sources

The collectors wired into Discover.

Group	Collectors
Research & Community	Hacker News, Reddit, Hugging Face, arXiv, Product Hunt, LMSYS / LiveBench, There's An AI For That, Exploding Topics
Industry	Anthropic Blog, OpenAI Blog, DeepMind, Artificial Analysis, generic RSS/Atom
Trends	Google Trends, PyPI Trends, Web Search (MCP), Website Deep Scrape

Each collector has a frequency (how often it runs), a module allowlist (which of Discover/Listen/Compare can use it), and a per-instance config (query terms, subreddits, categories…). You can collect now, pre-filter on demand, or delete the instance from the Sources page.

Highlights

Three things that matter day-to-day.

Curated, not a firehose

Every item is pre-filtered and labeled before it reaches the report — you see summaries + categories, not raw RSS.

Report history

Every generation is stored. Compare this week's highlights with last week's without running anything again.

Chat the feed

Each module has its own chat scoped to the filtered items — no cross-product contamination.

See PMOS end-to-end.

A 20-minute demo walks through Discover → Aggregate → PRD → GitHub issue, scoped to your product.

Get a demo Read the docs