Clawnection MVP (Prototype 1)

Clawnection is a romance-first, agentic matchmaking prototype.

In this MVP, each person is represented by a personal matchmaking agent. Before two humans decide to meet, their agents run a bounded virtual date protocol and generate a recommendation (meet, maybe, or not recommended).

The key principle is human-in-the-loop decision making:

• agents provide structure and recommendations
• humans remain in control of whether to proceed in real life

What this MVP includes

• Landing page at / with product framing and clear CTAs.
• Onboarding flow at /onboarding for building a lightweight romantic profile.
• Seeded romantic counterpart profiles in local TypeScript data.
• Shared domain types for profiles, rounds, recommendations, concerns, and match results.
• A common agent adapter interface plus:
  • HostedAgentAdapter
  • MockExternalAgentAdapter (simulates bring-your-own-agent workflows)
• Deterministic virtual-date simulation engine with six rounds + closing assessment.
• Demo flow at /demo to run simulations.
• Results page at /results with score, strengths, concerns, recommendation, and transcript summary.

Tech + architecture constraints

This prototype intentionally avoids backend complexity:

• No authentication
• No database
• No Supabase
• No OpenAI API integration yet
• No OpenClaw integration yet
• No external matchmaking dependencies

State is local-only using:

• local files
• TypeScript modules
• localStorage

Run locally

npm install
npm run dev

Open http://localhost:3000.

What is mocked vs future work

Mocked in this MVP

• Agent behavior is deterministic and rule-based (no LLM calls).
• External agent support is represented by a mock adapter implementation.

Future extensions

• Plug hosted adapter into real model inference.
• Add real bring-your-own-agent protocol adapters.
• Add persistence, user accounts, and match history.
• Add OpenClaw support and richer multi-agent orchestration.

For a condensed roadmap, see PROJECT_SPEC.md.