Skip to content
Compesight
Start free

What we've shipped.

Newest first. We'll keep this honest, including the parts that took longer than they should have.

  1. 0.9

    Commercial

    BillingLimits
    • Plans, Paddle checkout and workspace subscriptions.
    • Plan-bound crawl intervals and competitor limits, enforced server-side.
    • LLM budget enforcement: a soft warning at 80%, and a hard stop that queues classification rather than dropping it.
    • Snapshot retention. Raw payloads drop after 30 days, and the latest per source is kept.
    • Google Sheets and Docs export.
  2. 0.8

    Cited Q&A

    Q&A
    • Ask questions in natural language across a competitor's accumulated findings.
    • Every answer cites the findings it drew from.
    • When monitoring holds nothing relevant, the answer says so instead of drawing on the model's own knowledge.
  3. 0.7

    Seven more source types

    Sources
    • RSS/blog, Reddit, Hacker News, Product Hunt, YouTube, GitHub, App Store and Google Play.
    • Sources became adapters behind a registry, so crawling and diffing no longer branch on type.
    • Source discovery probes a domain for likely feeds, repos and channels. Nothing is monitored until you confirm it.
  4. 0.6

    Alert routing and feedback

    Alerts
    • Routing rules: (change type, minimum urgency) → channels, per workspace with per-competitor overrides.
    • Email, Slack and webhook channels.
    • Mutes with optional auto-unmute, and cross-source deduplication so one change is one alert.
    • Thumbs up/down and corrections on every finding.
  5. 0.5

    Battlecards

    Battlecards
    • A per-competitor battlecard covering positioning, strengths, weaknesses and pricing tiers, maintained as findings land.
    • Section-level provenance: each section shows when it last changed and links to the finding that changed it.
  6. 0.4

    Classification and urgency

    Pipeline
    • Changes classified by type and urgency, with model output validated against a schema before storage.
    • Malformed output is preserved as classification_failed rather than dropped or invented.
  7. 0.3

    Source health

    Reliability
    • Healthy, degraded, failing and blocked as distinct states throughout the product.
    • A source that can't be crawled never renders as "no changes found".
  8. 0.2

    Crawl, extract, diff

    Pipeline
    • Playwright rendering, structured extraction with an LLM fallback, and field-level diffing.
    • Baseline capture on adding a competitor, so the first crawl produces a battlecard rather than an empty screen.
    • Polite crawling: robots.txt, rate limits and backoff.