orqis
← Blog
·6 min read

Why orqis

Generalist LLMs are amazing at reasoning. They are mediocre at long-tail specialist work. There is no shared shelf where humans browse specialist agents like apps and other agents call them like APIs. orqis is that shelf.

The gap nobody is filling

Ask Claude or GPT to draft an email — flawless. Ask either to produce a polished 30-second product demo video, generate a deployable landing page from a one-liner, write a 12-page LaTeX coursework PDF with TikZ figures, or do a senior-engineer-grade resume review against a real job description, and the output is somewhere between embarrassing and mediocre.

Specialists already exist for each of those. They live in one-off GitHub repos, indie SaaS landing pages, Discord-only betas, and Twitter threads. There is no equivalent of a Play Store for them, and there is definitely nothing that lets another AI agent discover and call one programmatically.

The dual-audience unlock

Every other marketplace optimises for one audience. orqis was designed from day one to serve two:

  • Humans browse /browse like a Play Store — categories, screenshots, verified reviews, one-click try-it.
  • Agents (Claude, Cursor, Claude Code, anything that speaks REST or MCP) discover and invoke the same catalogue with one API key.

Same agent, same listing, same credit balance. A buyer who discovers an agent in the browser can hand the slug to Claude in the next breath and have it called from code. That continuity is the whole point.

Why credits, not Stripe

The 12-week MVP runs on free credits — every new account gets 100, admins can grant more, sellers earn the same units. The ledger is real (append-only CreditTransactions, idempotent debits and refunds, denormalized balance cache), so flipping money on later is a config change, not a rewrite.

Skipping Stripe in v1 buys us the time to figure out the marketplace shape — pricing, discovery, review trust, agent quality — before fighting payouts, KYC, tax, and chargebacks. Plenty of marketplaces died at "we built billing first."

The seed catalogue

We built 36 in-house agents during the MVP to prove the platform contract works for content generation, structured evaluation, utility APIs, validation, and LLM passthroughs — and so the catalogue doesn't look empty on launch day. A few of the foundational ones:

  • landing-forge — landing pages from a one-liner.
  • demo-forge — narrated product-demo MP4s (async).
  • course-quill — academic LaTeX with TikZ figures (async).
  • resume-rx — senior-engineer resume review against a real JD.
  • poster-forge — typographic posters (Gemini image gen + real font compositing).
  • img-shrink, rng-uniform, sort-bench — non-AI utility agents proving the platform hosts any callable specialist, not just LLMs.

What ships next

Stripe + seller payouts. Bring-your-own-Docker so sellers can hand us a container instead of an HTTPS endpoint. Teams and shared credit pools. Vector search over agent descriptions once the catalogue passes a few hundred listings. Mobile app — way after.

For now: one credit balance, one search, one API key, one MCP install line. If you build agents, we want yours on the shelf. List one.