Using meith

Working with agents

In meith, agents don't just chat — they call real tools to build your web app, with every step shown inline.

Agents work in your project's context

Open an agent chat tab inside a workspace and the agent operates within that project. It can read and edit your app's code, run the dev server, drive an embedded browser to check the live preview, and inspect dev-server logs — all through the same shared tool registry the rest of the app uses.

Provider-agnostic by design

meith doesn't lock you into one AI provider. The agent runtime uses an adapter interface, and currently includes:

a built-in mock adapter for local testing without any external model,
an ACP subprocess adapter that connects to real external agents over the Agent Client Protocol.

The ACPpath lets an agent use meith's tools without forcing the app to depend on a particular AI vendor or SDK.

You approve what matters

Agents do not get unrestricted access. Read-only work runs without interruption, but anything that changes your machine pauses for approval:

file writes,
browser control,
process starts,
network requests and destructive actions.

Edits are reviewable

When an agent edits a file, the change lands as an inline diff with a gutter marker, so you can see exactly what was rewritten — and undo it.

Learn more

For how permissions and tool calls work under the hood, see Tools & permissions. Developers integrating agents should read the Agent runtime reference.