Vercel Labs releases Emulate for stateful GitHub, Vercel, and Google API tests

• Vercel Labs has released emulate, a local service emulator for GitHub, Vercel, and Google that aims to replace brittle mocks with "stateful" API behavior for integration testing Emulate launch.
• The pitch is deterministic external-service testing for agents, CI, and "no-network sandboxes," with support for auth flows such as "Sign in with Google" without hitting Google itself Google auth demo Repo summary.
• The project ships as a CLI and skill install, and the linked repo describes zero-config local startup plus seedable configs for users, repos, orgs, teams, and OAuth clients Repo summary.
• The launch also fits a broader pattern around agent-facing skills: LlamaIndex separately shipped a one-line LlamaParse skill for turning complex PDFs into markdown that agents can consume LlamaParse skill.

Vercel Labs is positioning emulate as a production-fidelity alternative to mocks for third-party integrations. In the launch post, ctatedev describes it as a service emulator that makes external integrations "easy to test, stable to run, and predictable" for agents, CI, or any workflow where determinism matters Emulate launch.

The initial service set is GitHub, Vercel, and Google Service list. According to the linked GitHub repo, the tool can run multiple services locally, expose service-specific configs, and load seed data from YAML or JSON. The repo summary says developers can start everything with npx emulate, target specific services, customize ports, and use a programmatic API in test runners such as Vitest and Jest Repo summary.

The most concrete demo is Google auth. Vercel's post shows "Sign in with Google" working "without actually signing into Google," which is a practical fix for one of the hardest parts of end-to-end testing: reproducing third-party login and OAuth behavior without flaky network dependencies Google auth demo.

That matters more as agent tooling starts depending on installable skills and external services. Jerry Liu's LlamaParse skill post shows the same one-line skills pattern being used to give agents access to document parsing for "dense tables, unlabeled charts, messy handwriting and more." Put together, the pattern is clear: if agents are going to call real-world tools, developers also need local, predictable stand-ins for auth and API flows during development and CI Emulate launch LlamaParse skill.