features

Built for serious agent work.

A detailed look at what CloudAgent handles for you — from the first SSH connection to the merged pull request.

# 01 · any-terminal access

Use your own terminal.

IDE integrations can come later; the terminal is the source of truth.

  • SSH-first access
  • Works with common desktop terminals
  • Works from remote/mobile SSH clients
  • No required IDE extension
  • Persistent terminal preferences
  • Browser fallback terminal later
Terminal.app
iTerm2
Ghostty
Warp
Windows Terminal
VS Code terminal
iPad SSH apps
Browser (later)

# 02 · persistent background work

Close your laptop. The task keeps running.

  • Sessions survive disconnects
  • Tasks continue while you are offline
  • Automatic reattach to your last active task
  • Durable logs and transcripts
  • Task state visible from the dashboard
  • Notifications for completion or required input

# 03 · agent-ready environment

Major coding agents, maintained and ready.

Agent availability is labeled honestly: planned beta support until each one is verified in the runtime.

  • Claude Code, Codex CLI, Gemini CLI
  • Aider and OpenCode
  • Common package managers and runtimes
  • Version-pinned agent images
  • Agent health checks

# 04 · repo boot & personalization

Your repo, ready before your coffee.

  • GitHub repo clone
  • Devcontainer detection
  • Dotfiles sync and shell preferences
  • SSH/Git config
  • Package and runtime caches
  • Docker Compose and preview URLs later

# before

install runtimes → install agent CLIs → copy secrets → configure auth → start tmux → hope the laptop stays awake

# after

ssh dev@cloudagentrun task

# 05 · billing & model control

Model usage is visible by task.

  • One credit balance
  • Model-specific rates shown before you run
  • Per-task token tracking
  • Input / output / cached-token breakdown
  • Hard spend caps and low-balance warnings
  • Purchase history and admin-granted beta credits
  • Bring your own keys later

# 06 · safe autonomy

Autonomy with boundaries.

  • Per-task branch/worktree isolation
  • Scoped secrets
  • Approval policies and dangerous-command prompts
  • Audit trail
  • Spending limits
  • Credential redaction in logs
Read onlyEdit filesRun testsInstall packagesAccess networkAccess secretsPush branchOpen PRDeploy

Autonomy levels are granted per task, not per account — an agent that only needs to read never gets push access.

# 07 · review & handoff

Every task ends in something reviewable.

  • Changed files and diff summary
  • Test results and logs
  • Token cost per task
  • Agent transcript
  • PR-ready branch
  • Follow-up prompt context

# review panel — illustrative

[done] task t-0142 · fix flaky auth tests

+86 −23 · 4 files changed

tests: 212 passed, 0 failed

cost: 2.31 credits (claude-sonnet-5)

src/auth/token_refresh.py

tests/auth/test_refresh.py

tests/conftest.py

PR #142 → ready for review

Request access to the private beta.

We are inviting developers and small teams who already use AI coding agents and want an always-on cloud environment for serious work.