Turn every feature you have ever built into a revenue stream.
Any capability from any app — built or only documented — becomes a swappable, sellable, bundleable unit. Customers pay in credits and receive results in minutes, backed by an honest SLA.
Not a storefront. A liquidation layer for software assets.
Most SaaS products are sold as monoliths: one subscription, one app, one feature set. AgentCraft Marketplace breaks that mold — every atomic capability (a single input → single output transformation) becomes an independently sellable asset, priced by its AI complexity.
Built apps become mines of sellable services
A geo-scraping pipeline or a domain health check inside an existing product starts earning as a standalone service overnight.
Documented studies become pre-validated inventory
A PDF describing an idea you never built is parsed, classified, priced and queued — ready to publish when you say go.
Market signals suggest the next service to build
Demand from Fiverr, Upwork and Reddit — filtered by a two-signal rule — tells the catalog where to grow next.
A domain health check extracted from Scout, a product URL analysis from Publisher, a recitation evaluator from a documented study — all live side by side in one catalog, each priced by its AI complexity.
Four engines continuously feed one shared knowledge base — and every new service discovered makes the next discovery cheaper, through structural pattern reuse.
Four engines. One shared knowledge base.
The heart of the system: four engines operate on one knowledge base (PostgreSQL 16 + pgvector), each feeding the others in a continuous self-improvement loop.
Extraction
Upload a project document (PDF / DOCX / TXT, Arabic OCR included) or paste raw text. The engine chunks long documents, detects the language, extracts atomic services and classifies each on three axes — service type, AI layer, structural pattern — with confidence and readiness scores. Runs asynchronously; results merged and deduplicated.
Combination
Discovers bundles across published services in two phases — cheap rules first: pipeline chains, shared-pattern upsell and audience bundles; then LLM evaluation of real synergy with a proposed name and justification. Admins can also select 2–5 services manually for an instant bundle with a live price preview.
Market Intelligence
Monitors external demand — Fiverr, Upwork, Reddit, G2 — to surface two opportunity types: proven high demand, and real rarity (scarcity plus documented complaints, per a two-signal rule). A scarcity signal without demand is flagged as a potentially dead market and deprioritized.
Continuous Development
Every admin rejection, refund reason and support ticket is captured as feedback with an extracted topic — slow, low accuracy, high price, feature request. Periodic sweeps recommend a per-service action: upgrade, extend, fix, reprice or retire.
One shared knowledge base — PostgreSQL 16 + pgvector. Every service carries a 3-axis classification, an embedding vector and a lifecycle state across 13 marketplace tables. Vector cosine search powers pattern matching, so duplicate logic under different names is rejected before it enters the catalog.
15 recurring structures. Built once, sold many times.
A first-class asset: fifteen pre-seeded logical structures, reusable across domains. Two services from completely different projects can share the same pattern. New patterns are matched against this library via embeddings before acceptance.
| Pattern | What it does | Real examples |
|---|---|---|
| strict_ai_evaluator | Rubric-based scoring with a detailed report | Quran recitation · mock immigration interview |
| url_to_structured_data | URL → structured analysis | Publisher product analysis · Idea Discovery |
| formal_text_generator | Formal / marketing copy in a specific context | Publisher articles · immigration cover letters |
| semantic_matcher | Meaning-based matching (not keyword) | Immigration programs · SaaS idea matching |
| deterministic_scoring_engine | Weighted scoring on fixed signals, no LLM | Scout company quality · Idea Discovery |
| domain_health_checker | Pure technical check, zero AI | Scout site checker |
| geo_scraping_pipeline | Structured scraping by geographic subdivision | Scout · Industrial Marketer (81 provinces) |
| outreach_automation | Outreach queue with scheduled follow-up | Scout · Industrial Marketer |
| intent_extraction | Free-text request → structured search criteria | Smart Traveler · tool RFQ agent |
| comparison_report_generator | Table + rationale across candidates | Smart Traveler · tool RFQ agent |
| personalized_roi_estimator | Personalized ROI for the user's own case | Tool RFQ agent |
| policy_grounded_responder | Auto-reply grounded in a policy / knowledge base | SRI review replies · policy email replies |
| meta_agent_provisioning | Provisions another custom agent on demand | Build me a custom agent |
| form_to_generated_website | A filled form → a full website | Digital Identity Factory |
| directory_listing_setup | Register / set up presence on external directories | Google Business · B2B platforms |
Pattern identifiers stay in English by design — they are canonical code-level names shared across the platform.
Priced by AI layer. Never randomly.
Customers buy credits — by subscription or one-off — and each order deducts credits by how much intelligence the service actually uses.
Pure technical checks with near-zero cost — domain health, URL status.
Rules and weighted scoring — possibly a paid data fetch, never an LLM call.
A full LLM run per execution — analysis, drafting, evaluation.
Five people who already need this
The solo founder with too many apps
You have built seven products; each contains features people would pay for — but building a checkout per feature is uneconomic. Marketplace extracts them once and sells them forever: a geo-scraping pipeline becomes a $5 standalone service overnight.
The agency serving many clients
Competitor analysis for client A, an ROI calculator for client B, outreach automation for client C. Instead of three subscriptions, one Pro bundle combines three atomic services with a 12.5% discount — one credit pool, three results, delivered in minutes.
The researcher with a documented idea
A nine-page PDF describing a price-monitor agent you never built. Upload it: the extraction engine reads it — Arabic OCR included — identifies five atomic services, classifies each on three axes, prices them, and queues them ready to publish when you say go.
The developer buying building blocks
A SaaS founder needs intent extraction and comparison reports for a travel app — and discovers both already exist as atomic services sharing one structural pattern, built once for a completely different domain. Buy the bundle and ship.
The market-driven builder
Market intelligence reports forty-plus complaints on Reddit, an $80 average on Fiverr, and no clean solution — a real-rarity signal. Convert it into a source project; the extraction engine drafts the services; the catalog grows with proven demand.
Where this pattern already works
The extract → classify → match → sell pipeline is domain-agnostic. Four domains were researched and held up under scrutiny:
API marketplaces
Platforms like API Hub (formerly RapidAPI) already list atomic capabilities with tiered pricing over a unified catalog. What they lack is the extraction half — nobody proposes new listings from a provider's existing surface automatically. That is exactly what this engine adds.
Internal developer platforms
Enterprise catalogs already fight teams rebuilding the same microservice under different names — the exact problem the 15-pattern library solves, scoped to one company. A platform lead with dozens of scattered APIs wants precisely this extraction-and-matching logic.
No-code template marketplaces
The closest match found. Zapier, Make and n8n marketplaces already sell packaged automations as standalone products — extract once from work you already did, sell forever — with pricing that maps directly onto AI-layer tiering.
University technology transfer offices
A formal process for turning documented, uncommercialized research into licensable revenue — the researcher use case at institutional scale. Today it is high-friction and negotiated; that is precisely the friction an instant, self-serve, atomic catalog removes.
Engineering discipline behind the storefront
Atomic credit deduction
Balance is checked and deducted in a single database transaction. No half-paid orders — ever.
Honest SLA enforcement
Per-service delivery type, a 5-minute early-warning sweep, and breached orders surfaced in red for immediate action.
5-layer disaster recovery
Daily database, environment and upload backups; one-command restore; monitoring with alerts.
Validation everywhere
Pydantic v2 schemas tolerate malformed LLM responses — strings-as-objects, bad scores — without breaking the pipeline.
Graceful OCR degradation
If OCR is unavailable, direct-extraction results are returned — a soft fallback, never a hard failure.
Self-improving loop
Every rejection reason is captured and reused, so future AI suggestions keep getting better.
Where the platform stands today
Foundation
Shipped13 marketplace tables + AI / embedding abstraction + auth, billing and disaster recovery inherited from the platform template.
Catalog + orders
ShippedAI-assisted service authoring, public catalog, credit checkout and the SLA monitor.
Extraction engine
ShippedUpload / paste documents → candidate services, OCR for scanned Arabic PDFs, embeddings per service.
Combination engine
ShippedAuto-discovered bundles (three criteria + AI evaluation) and a manual bundle builder in the review queue.
Custom request flow
ScaffoldedCustomer free-text → intent extraction → instant bundle quote. Tables are ready.
Market intel + continuous dev
ScaffoldedExternal demand monitoring and the self-improvement loop. Tables are ready; scraper integration is next.
Ready to buy your first result?
Start with a free plan. No credit card required.