Veetlyn

Sell where clients ask AI.

Veetlyn puts a seller's services and products inside Claude and ChatGPT. A client's AI finds the offer, asks questions and requests a quote. The client pays with PayPal on Veetlyn. Agents work within the permissions and business policies assigned to them. In the alpha, buyers approve checkout in PayPal; team payouts follow approval and automation settings.

Website · Network and plugin · License

Status: alpha. Payments run on the PayPal sandbox while the product is in alpha. veetlyn.com is open to the launch waitlist; the staging deployment is open for testing.

What it does

• Quotes and projects. Clients request work; a team's freelancers ask questions and quote; the client accepts one quote, pays a deposit, follows the delivery and pays the balance.
• PayPal, done properly. Orders are created on the server, every payment is approved by the payer in PayPal, webhooks are verified, and refunds and disputes pause what they should.
• Team payouts. Payout rules split each payment between members; a person or a rule approves each payout before PayPal sends it.
• Veetlyn Network. One connector (https://api.veetlyn.com/mcp) lets Claude and ChatGPT search every listed seller, request quotes and follow projects. The alpha connector prepares purchases for buyer approval in PayPal.
• Your own agents, with limits. Every action goes through one registry with roles and per-action agent policy: free, needs a person's confirmation, or never.
• Found in search and in AI. Seller pages on <seller>.veetlyn.app with structured data, open to AI crawlers, plus guided visibility work.

Use it from your AI assistant

Download the plugin from veetlyn.com/network, or add the connector by hand:

• Claude: Customize → Plugins → Upload plugin, or Settings → Connectors → Add custom connector with https://api.veetlyn.com/mcp.
• ChatGPT: Plugins → Upload, or turn on developer mode and add the same address.

The plugin source is in plugins/veetlyn.

How it is built

Claude / ChatGPT ──MCP──┐
                        ▼
browser ──► web Worker (vinext) ──service binding──► api Worker (Hono, tRPC, MCP)
             veetlyn.com, dash.,                       api.veetlyn.com
             *.veetlyn.app                                   │
                                                             ├─► Postgres (Hyperdrive, Drizzle)
                                                             ├─► R2, KV, Queues
                                                             └─► PayPal, Resend, Discord

• Runtime: Cloudflare Workers. Infrastructure as code with Alchemy in packages/infra/alchemy.run.ts.
• Web: vinext (Next.js on Vite), Tailwind, shadcn/ui primitives in packages/ui.
• API: Hono, tRPC and an MCP server; Better Auth for sign-in and OAuth for the connector.
• Data: PostgreSQL through Hyperdrive, Drizzle ORM and migrations.
• Config: Varlock schemas (.env.schema) in every app; no secret lives in the repository.

Repository layout

apps/
  web/            website, dashboard and seller pages (vinext, one Worker)
  server/         api Worker: tRPC, MCP, webhooks, queues
packages/
  api/            use cases and the action registry
  auth/           Better Auth setup
  core/           domain rules shared by both Workers
  db/             schema, repositories, migrations
  paypal/         PayPal client
  email/          email templates and delivery
  ui/             design tokens and shared components
  infra/          Alchemy stages, hosts and deploy scripts
  oss/            standalone MIT packages (rules engine, outbox, scheduler)
plugins/veetlyn/  the Claude and ChatGPT plugin

Run it locally

You need Node 24, pnpm 10 and Docker.

pnpm install                       # also generates the typed env accessors

# Postgres for development
docker run -d --name veetlyn-pg -p 5432:5432 -e POSTGRES_PASSWORD=postgres postgres:17
docker exec veetlyn-pg createdb -U postgres veetlyn_dev

Fill the ignored env files from their schemas: apps/server/.env (see apps/server/.env.schema) and packages/infra/.env. Set DEV_DATABASE_URL (in packages/infra/.env) to the database above. PayPal values come from a PayPal developer sandbox app.

pnpm db:migrate:dev                # migrate the dev database
pnpm dev                           # site on localhost:3001, dashboard on dash.localhost:3001

Alchemy emulates R2, KV and Queues locally.

Checks

pnpm check-types
pnpm lint          # Oxlint with the anti-slop plugin, plus the brand and OSS-boundary checks
pnpm test          # Vitest; needs TEST_DATABASE_URL in .env.test.local, on a database whose name contains "test"

CI runs all three, a secret scan and the release builds on every pull request.

AI plugin builds

Built plugin zips are never committed.

• pnpm plugin:zip and pnpm plugin:zip:staging write the Claude and ChatGPT zips to dist/plugin/.
• The AI plugin workflow builds both stages on every change to plugins/ and keeps them as the ai-plugin-zips artifact.
• Each deploy builds the downloads for its own stage into the website, served at /plugin/claude.zip and /plugin/chatgpt.zip.

Deploying your own

Two stages are deployed from GitHub Actions: staging (after CI passes on dev) and production (an owner-approved dispatch from main, with verified staging evidence for the same tree).

Hosts are derived from two zones in packages/infra/deploy-hosts.ts: the platform zone (site, dash., api.) and the seller zone (*. seller pages). Only production is indexed by search engines; other stages send noindex.

Deploys use CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID. The token needs, on the account: Workers Scripts, Workers R2 Storage, Workers KV Storage, Queues, Hyperdrive and Account API Tokens (Edit), Account Settings (Read); on both zones: DNS, Workers Routes and Dynamic URL Redirects (Edit), Zone (Read).

Stage values live in ignored files (packages/infra/.env, packages/infra/.env.<stage>.local) and in the GitHub environment of the same name; pnpm -F @jendal-workspace/infra github-env <stage> copies them and prints names only. PUBLIC_ACCESS=waitlist closes sign-up and sign-in behind a launch list; open opens them.

Branch protection for main is saved in .github/main-protection.json; reapply it with gh api --method PUT repos/veetlyn/veetlyn/branches/main/protection --input .github/main-protection.json.

Documentation

• docs/environments.md — local, staging and production: what each stage is for, who can sign up, and the demo/sandbox rules
• docs/user-guide.md — using Veetlyn as a seller (services, payouts) or a client (requests, quotes, payments)
• docs/ai-plugin.md — connecting Claude and ChatGPT, what to ask, and the rules the connector cannot break
• docs/development.md — environment variables, the test database, migrations, code rules and project workflow
• CONTRIBUTING.md — how to open issues and pull requests

Contributing

• Open pull requests against dev, the default branch. main only receives release merges.
• Use Conventional Commits.
• Keep pnpm check-types, pnpm lint and pnpm test green; lint must stay at zero findings.
• Read CLAUDE.md for the code rules (money in integer minor units, untrusted input decoded with zod, no agent moves money without a person). They apply to people too.
• Product designs and plans are kept in a separate private repository. Open an issue to discuss a change before building it.

Security

Please do not open public issues for vulnerabilities. Report them privately to the maintainer through GitHub (@Mouhamed-B). The app only touches the PayPal sandbox today.

License

Veetlyn is open core:

• Everything outside ee/ is AGPL-3.0-only, or available under a commercial agreement.
• ee/ holds the enterprise features under the Veetlyn Enterprise License: free to develop and test, a commercial agreement for production.
• Standalone packages in packages/oss/ are MIT.

Details are in LICENSING.md. The Veetlyn name and logo are not licensed by any of these licenses.

If this project helps you, you can sponsor it.