OPENREAP.
An x402-native marketplace where AI agents pay other agents in USDC for queries. Your expertise gets wrapped as a skill file, deployed by OpenReap, then earns while you sleep. The brief was: make 'agents pay agents' feel like a normal weekday.
- ROLE
- Sole designer · brand → product
- CLIENT
- OPENREAP
- YEAR
- 2025
- SCOPE
- LANDING · PRODUCT · DATA-VIZ
- LIVE
- openreap.vercel.app ↗
- FIGMA
- ↗ source file
YOUR EXPERTISEEARNS WHILEYOU SLEEP.
OpenReap had a strange brief: design a marketplace where the buyers aren't humans. AI agents browse, query, and pay other AI agents — but humans (the people uploading their expertise) still need to feel the product belongs to them. Two audiences, opposite expectations, one landing page.
I leaned into that tension instead of hiding it. The visual ground had to read warm and editorial for the humans — closer to a printed magazine than a SaaS dashboard — while the underlying structure stayed strict and schema-friendly for the agents reading the same page. The cream parchment palette and the ember accent did most of that emotional work.
Cream parchment ground instead of the default dark void. The ember orange appears only on 'earn' actions — never on navigation, never on background. Restraint is the brand here.
Reframed the unit of work from 'a service you sell' to 'a file you upload once.' That single noun-shift drove the whole landing flow — Upload → Deploy → Earn — three cards, each carrying one big number, no buried CTAs.
Agent profile pages had to be browseable by humans but parseable by other agents. I built every card as a self-contained data row — Lighthouse-clean, schema-rich, copy that reads as a sentence either way.
Unique screens shipped — landing, marketplace, agent profile, templates, pricing.
Core flows mapped end-to-end: upload, browse-as-agent, browse-as-human.
Design rounds with the founding team before commit. Most decisions stuck on round one.
The first thing Saksham showed us was a colour swatch and a sentence. He said 'we're going parchment'. We thought he was joking. He wasn't, and now I can't imagine the brand any other way.
RAVEN.
Decentralised LLM memory infrastructure. Three lines of code give an agent a persistent, retrievable memory that survives across sessions and stacks. Stored on Walrus, indexed for vectors, priced in tiers — the design had to make 'memory' feel like a physical product.