Personal news app.

The problem

Busy professionals still want to be informed but can no longer stomach how it feels. Every session leaves them worse off — alarm and outrage framing dominate even mundane news, the same story is repackaged across outlets, and genuine new information is buried under a noise tax.

They also lose continuity: caring about a story last week means digging across multiple apps to find what changed. Existing AI aggregators don't solve this — the target persona finds even the category leader, Particle, overwhelming. Mood damage is the highest-frequency, highest-severity pain, felt every session.

The solution

Canopy Brief is a calm, AI-augmented personal news app. It clusters coverage of the same event from multiple sources and produces one neutral, wire-service-style summary with per-claim provenance, so every claim links back to a source the reader can verify.

It is calm by design, not a feature-rich aggregator: a 5–7 card home feed that terminates in "you're caught up," no infinite scroll, and a "what changed since last visit" diff for followed stories. Structured contestability lets users flag a specific claim, a wrong cluster merge, or a hallucinated citation — turning mistakes into labeled data that feeds the evaluation loop.

How it works

The summarizer runs as a background batch feature, not a chat surface. A versioned master system prompt casts the model as a wire-service news editor; the user message is a machine-generated JSON payload of a cluster's articles. Output is strict, schema-validated JSON with headline, summary paragraphs, per-claim source attribution, and disagreements — enabling the tap-to-source provenance in the UI.

Hard rules enforce grounding (every claim tied to a supplied source), a verbatim cap to limit copying, inline uncertainty language, cross-source synthesis, and a defamation guardrail. Architecture is two-tier: Gemini 2.5 Flash-Lite handles the bulk of clean clusters cheaply with context caching, escalating only validation failures to gpt-5.4-mini via the Batch API. Single-article clusters bypass the LLM with a deterministic attributed lead. A programmatic gate blocks hallucinations, schema errors, over-length output, and verbatim overlap before anything reaches users.

Who it's for

The product is B2C, built for one persona: the busy professional, late 20s to mid 40s, who currently juggles 3+ news channels, has 5–20 minutes a day, and ends each week mood-damaged and still behind on what matters. They've tried muting, deleting apps, and digest newsletters — none stuck, and dropping out isn't an option.

Revenue is freemium subscription. The free tier delivers personalized clusters, provenance, and limited AI chat; the paid tier (target $5–10/month) adds custom sources, higher chat caps, and push or email summaries — honest pricing for a specific professional segment, not mass-market scale.

Why it matters

News faces low willingness to pay (~18% of users across 20 wealthy markets paid for online news last year), active publisher copyright pushback, and commoditizing AI capabilities. But awareness of doomscroll and mood damage is creating real demand for calmer products, and multilingual markets are severely under-served by English-first aggregators.

Now a pre-seed, pre-launch startup, the working assumption is the consumer AI news subcategory grows at ~25–30% CAGR off a small base. The differentiators are deliberate: structured per-claim contestability as a labeled-data generator, a conservative copyright posture that avoids paid-licensing dependencies, and calm-by-design as the core promise rather than a mode.

At a glance

Project

Personal news app

Built by

Sofia Kolchanova

One-liner

This project is a personal news app for busy professionals who want to stay informed without emotional overload.