# Republic.dev

Discover the companies building the physical world.

AI Builder, Devekko.dev. Four-week AI builder sprint, October–November 2026. In progress.

Snapshot: October 29, 2026. Directory live · Marketplace prototype.

Demo: https://republic.dev

A directory and multisided marketplace prototype connecting industrial companies, buyers, suppliers, and the people researching them.

## Product intent

Industrial discovery often starts with scattered company websites, news, and supplier lists. Republic brings that information into a directory organized around industries and capabilities. The product hypothesis is that a useful company profile can become the starting point for a buyer, supplier, service provider, or researcher to make a relevant connection.

The demo explores reindustrialization and physical AI as a market: the businesses, products, services, and relationships behind manufacturing and automation. Its scope is discovery and marketplace software.

## What the demo shows

The public site includes industry browsing, researched business profiles, capital research, reported deals, and product and service discovery. Visitors can suggest a business and explore a conversational guide. Directory profiles carry source and verification context.

The marketplace remains a prototype. Catalogue presentation and simulated commerce do not establish available inventory, real payment processing, or fulfillment. A researched company listing does not imply a partnership or a verified business relationship.

## Product specification

**Primary users:** industrial buyers, company operators, suppliers, service providers, and researchers comparing a market.

**Core workflow:** select an industry or capability, inspect a company and its sources, compare relevant offerings, and prepare a qualified connection or inquiry. A company operator should be able to propose corrections through a reviewed process.

The proposed next release should support:

- Company, capability, product, service, and relationship records with stable identities and provenance.
- Search and filters that distinguish a company's stated capabilities from inferred classifications.
- Separate buyer, seller, and operator permissions, with review before a submitted change becomes public.
- Inquiry routing with an explicit recipient and a record of delivery state.
- Marketplace status that clearly distinguishes previews, simulations, and transactable offers.

Acceptance means a visitor can follow a profile back to its evidence, a reviewer can resolve a correction, and a tenant cannot access another company's private records. Any future transaction flow needs its own payment, inventory, and fulfillment acceptance checks.

## Technology

Implementation evidence includes TypeScript and Hono for the public application, Cloudflare Workers, PlanetScale Postgres, and a Python service layer with Pydantic contracts. Private container services support selected research and operator pilots. Their existence does not make every specified workflow a public feature.

The wider research architecture includes Pydantic AI, Prefect, Nemotron, and Brave API integrations. These components have different maturity levels across jobs. A production research adapter, a local agent experiment, and a recurring hosted workflow are separate capabilities that need separate evidence.

Codex and Claude belong to the sprint's development toolkit. They support implementation and review; release decisions and access policies remain the builder's responsibility.

## Research and development

The next experiments focus on discovery quality:

1. Compare capability-based search with broad industry categories using a reviewed set of buyer questions.
2. Test company matching against similarly named businesses, duplicate domains, and conflicting sources.
3. Assess how source dates and verification labels affect a user's confidence in making an inquiry.
4. Trace a submitted correction through review and publication, including rejected or incomplete evidence.

Research automation should prepare candidate records with citations and uncertainty labels. Publication requires review. Marketplace liquidity and the willingness of companies to participate remain hypotheses to test.

## Business measures to test

Track qualified inquiries per discovery session, time to find a relevant company, company-profile correction rate, and evidence freshness. For operators, measure review minutes per accepted record and research cost per usable profile. For a future marketplace, measure matched requests and repeat participation before treating catalogue traffic as commercial success.

These are proposed evaluation measures. Baselines and targets have not been published as sprint results.

## Sources and next release

The public experience was reviewed on October 29, 2026: [Republic.dev](https://republic.dev). This write-up also draws on project implementation and status records supplied in the builder's workspace. Private workflows were not exercised for this review.

The next release decision is whether a narrowly defined discovery-to-inquiry flow is useful enough to justify deeper marketplace operations.