Skip to content

Run bots in hosted Akeru Cloud sandboxes #428

Description

@linear-code

Problem

Running bots in the cloud today means signing up with a sandbox provider, copying API keys into Settings, and choosing between seven providers with different limits. Most users never get past that.

Goal

Hosted sandboxes through Akeru Cloud (beta) that a user sets up in one click: sign in, add credits, done. A bot set to Akeru Cloud runs its work in a hosted sandbox (4 CPU / 8 GB by default). Usage draws down the account's credits, and the sandbox stops when they run out. A short list of bring-your-own providers stays available: E2B, Vercel Sandbox, and Cloudflare Sandbox.

This builds on the Akeru Cloud foundation (AKR-129, PR #364).

Sub-issues

  • AKR-127: simplify sandbox providers to Local, E2B, Vercel, and a Cloudflare bridge, retire the old adapters, and migrate saved values
  • AKR-134: redesign Settings > Sandboxes around where bots run
  • AKR-135: provision Akeru Cloud sandboxes and bill credits
  • AKR-136: run bots in Akeru Cloud sandboxes from the environment
  • AKR-149: private per-bot workspaces on the environment machine

Acceptance criteria

  • Settings > Sandboxes offers Akeru Cloud as a one-click option
  • A bot set to Akeru Cloud runs its work in a hosted sandbox, usage draws down credits, and the sandbox stops when credits run out
  • Bring-your-own E2B, Vercel, and Cloudflare remain available
  • Every sub-issue above is done

How to verify

See each sub-issue. The integrated check is a real bot turn on an Akeru Cloud sandbox in an isolated dev environment against the staging cloud, plus Settings > Sandboxes screenshots at desktop and phone widths.

Context

Integration handoff, 2026-10-06: branch akeru-cloud/sandbox-integration (32 local commits on the cloud foundation #364, not pushed) covers AKR-127, AKR-134, AKR-135, and AKR-136.

  • PASS: 4,157 tests across cloud, server cloud, provider, routines, and settings, contracts, client-runtime, and web settings, roster, and palette, plus 64 mobile settings tests
  • PASS: typechecks for contracts, server, web, client-runtime, mobile, and cloud
  • PASS: lint on all 172 changed files
  • PASS: Settings > Sandboxes checked in a real browser against a fake cloud, at desktop and phone widths: unlinked, setup, linked with balance, low (minutes), empty, coming soon and available soon, offline, checkout success, a blocked non-web checkout URL, a setup error on a limited-scope device, and default reset on revoke
  • PASS: GPT-6.1 Sol reviews, two rounds each on billing and runtime and one on settings. All findings fixed with regression tests
  • NOT CHECKED: live hosted sandboxes and live checkout (staging credentials needed), native mobile on a device, and the Electron shell

Created with Claude Opus 5.5 in Claude Code.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions