@berthos/adapter-core

The DeployAdapter interface that every Berth deploy target implements. @berthos/adapter-e2b, @berthos/adapter-daytona and @berthos/adapter-k8s build on it, and the berth CLI talks only to this interface, never to a provider SDK.

Berth runs an AI agent's tools in a sandbox. Each tool declares what it may touch in a berth.yml manifest, and the Linux kernel enforces it.

npm install @berthos/adapter-core

You need this package only to write an adapter for a new provider.

Usage

import type { DeployAdapter } from "@berthos/adapter-core";

export function createMyAdapter(): DeployAdapter {
  return {
    name: "my-provider",
    upload: async (target) => pushImage(target.imageRef),        // -> { remoteImageRef }
    start: async (remoteImageRef, target) => bootInstance(remoteImageRef, target), // -> DeployHandle
    teardown: async (handle) => handle.stop(),
  };
}

| Member | What it does | |---|---| | upload(target) | Make target.imageRef available to the provider; returns { remoteImageRef } | | start(remoteImageRef, target) | Boot an instance; returns a DeployHandle (id, status(), streamLogs(), stop()) | | teardown(handle) | Stop and remove an instance | | list?, connect?, previewUrl?, rpcUrl?, pause?, resume?, fork?, snapshot? | Optional. Implement the ones your provider supports; the CLI falls back or reports the feature as unsupported otherwise |

DeployTarget is { imageRef, manifest, env?, region? }. withTimeout(), DEPLOY_CREATE_TIMEOUT_MS (5 min) and DEPLOY_READ_TIMEOUT_MS (30 s) are exported for bounding provider calls.

Docs

Kubernetes adapter reference (a worked example of the interface) · Repo