Sentiment Agent · WebRobot
Vertical AI agent · demo
Powered by the WebRobot tier

Sentiment analysis,
designed by an agent.

Describe what you want to understand. A vertical AI agent autonomously generates and validates a WebRobot ETL pipeline against the live stage catalog — a background Ray run, not a one-shot prompt — then runs it on the platform and charts the sentiment.

Demo — this runs on limited in-cluster resources, so the experience is illustrative on a small number of sources. A production tenant scales the crawl (more sources, higher volume) on dedicated capacity.

Insights

No sentiment data yet. Once a designed pipeline runs on the platform (scrape → analyze → persist), these charts fill in — polarity, the Plutchik emotion wheel, entities and trend.

Pay-as-you-go · indicative

What it costs, pay-as-you-go

Pure consumption — no subscription: you pay only what an analysis actually uses, in prepaid credits. Indicative, derived from the real costs (Claude tokens, proxy, compute, storage); billed from actual usage. 1 credit = €0.01.

  • AI inference (Claude tokens), per 1M in/out: Haiku ~138 cr (€1.38) / 690 cr (€6.90) · Sonnet ~415 cr (€4.15) / 2071 cr (€20.71)
  • Residential proxy (DataImpulse): ~138 cr (€1.38) / GB fetched
  • Compute (Hetzner, hourly per VM): ~5–10 cr (€0.05–0.10) / VM-hour
  • Storage (Hetzner Object Storage): ~98 cr (€0.98) / 100 GB-month
  • e.g. a sentiment analysis of 1,000 documents ≈ 415 cr (€4.15)
Acquisition infrastructure

Real data, from where opinion actually lives

The agent doesn’t guess where to look. It draws on a curated catalog of 1094 sources across 36 markets — news, forums, reviews, social, e-commerce and more — each with its fetch strategy already resolved, so it reads the right way the first time.

Browse the full source catalog — 1094 sources, 36 markets →
1094curated sources
36markets
16source categories
32/68direct vs browser fetch
Camoufox
A real Firefox, not a headless shell

Public pages that only render with JavaScript are read through Camoufox — a hardened Firefox driven by Playwright with a genuine fingerprint — so the page loads the way it does for a visitor.

DataImpulse
Residential egress

Requests leave through residential IPs rather than a datacenter range, so reading a public page looks like an ordinary home connection instead of a server farm.

Geo rotation
Country-matched exits

The exit IP is matched to the source’s country and rotates — an Italian forum is read from an Italian address — which keeps geo-specific pages reachable and rate limits comfortable.

Public & openMost sources — read directly, cleanly, at scale.
Anti-botAggressive protections are assessed per source: feasibility first, with any setup cost agreed up front.
Login / APIPlatforms that require it (e.g. social networks) are accessed via their official APIs, within their terms.
135 Italian sources · 67 opinion-bearing (forum · social · review · Q&A)
Disclaimer
  • WebRobot collects publicly available information and uses platforms’ official APIs where their terms require it. It does not perform scraping that breaches a site’s terms of use.
  • The figures above are an indicative snapshot of catalogued sources, not a commitment to collect from every one. Acquisition of any given source is not guaranteed.
  • Sources protected by strong anti-bot measures are evaluated case by case for technical and legal feasibility and may require a separately scoped and priced setup, agreed with you in advance.
  • Sentiment results are automated estimates for research and monitoring — not fact, financial, or legal advice.