Startup Deals Tracker

Structured intelligence feed of publicly-announced startup credits, free-tier programs, and SaaS discounts — aggregated from 18 official "for startups" pages and curated aggregators.

This is not a coupon scraper. It produces clean, schema-validated deal records — including provider, dollar value, eligibility, expiry, source URLs, and a confidence score — so you can rank, filter, and integrate them into founder-facing tools.

Why this exists

Founders waste hours hunting credits across vendor pages, aggregator sites, and Twitter threads. Most of what's out there is either dead, gated behind accelerators, or tagged with stale expiry dates. This actor gives you a single, filterable, freshly-verified feed.

What you get

A validated dataset where each row matches a strict Pydantic schema:

• Identity — provider, program_name, stable id slug
• Substance — deal_type (credits / free_period / percent_off / lifetime_deal / ...), headline, description
• Quantification — credit_value_usd, discount_percent, free_period_months, estimated_total_value_usd
• Eligibility — eligibility_tags (e.g. startup_seed, oss_maintainer, student), funding_cap_usd, requires_partner, geographic_restrictions
• Claim — application_url, public promo_code only, requires_application, expected_response_time_days
• Time — starts_at, expires_at, is_recurring
• Provenance — source_urls, source_type, first_seen_at, last_verified_at, verification_status, verification_method
• Categorization — confidence_score, category, tags

Sources

Official programs (12): Vercel for Startups, Vercel OSS, Render, Notion, Linear, AWS Activate, Microsoft for Startups Founders Hub, Google for Startups Cloud, NVIDIA Inception, Anthropic for Startups, MongoDB for Startups, GitHub for Startups.

Aggregators (4): CreditsGull, StartupPerks, AI Perks, AppSumo SaaS.

Newsletter / community (2): Lenny's Product Pass status, Product Hunt Stories.

How it works

1. Each source is fetched (fast HTTP for static pages, headless Chromium for JS-heavy sources).
2. Crawl4AI's BM25 content filter strips boilerplate and keeps only deal-relevant content.
3. Anthropic Claude Haiku extracts deals via a forced tool call producing structured JSON.
4. Every deal is validated against a strict Pydantic v2 schema; invalid records are discarded with a warning.
5. A 7-day content-hash cache skips the LLM call when a source hasn't changed (the main cost optimization).
6. Optional verify_freshness re-checks each source URL for "applications closed", "sold out", "waitlist" patterns.

Input

Field	Type	Default	Description
categories	string[]	[]	Filter by category (ai_tools, hosting, database, ...). Empty = no filter.
eligibility_filter	string[]	[]	Filter by eligibility tag (startup_seed, oss_maintainer, ...). Empty = no filter.
verify_freshness	boolean	false	Re-verify each source URL. Charges per verification-check.
min_confidence	number	0.7	Discard deals below this LLM confidence score.
force_refresh	boolean	false	Bypass the 7-day source content cache.
max_deals	integer	500	Hard cap on returned deals.

Pricing (pay-per-event)

Event	Price	When charged
deal-record	$0.005	One per validated deal pushed to the dataset
verification-check	$0.001	Per deal re-verified when verify_freshness=true

Notes

• Personalized promo codes are explicitly never extracted — only public terms.
• The actor tolerates source-level failures: if one source breaks, the others continue.
• Output is the default Apify dataset; consume via the standard Apify dataset API.

Development

For build, deploy, and local-run instructions, see ./DEVELOPMENT.md.

Author

Built and maintained by Bob @ CyberLink Security.