Bedrock at work, your account at home

The same CLI, two different feature sets. Which things to build where.

At work you run Claude Code through Amazon Bedrock. At home the mini signs in with a claude.ai account. The CLI is identical. The set of things it can reach is not, and knowing the split saves you from building an always-on agent at work and wondering why the channel never connects.

Works on both

Subagents, hooks, skills, CLAUDE.md and auto memory, plugins, MCP servers, checkpoints and /rewind, sandboxing, dynamic workflows, /batch, GitHub Actions. Cross-session messaging works on Bedrock between sessions on the same machine.

Home account only

Channels, so the Telegram bot on the Agents page. Routines. Remote Control from your phone. Claude Code on the web, mobile, and in Slack. The Desktop app, except through a separate third-party build. The Chrome integration. Computer use. /ultrareview and Code Review. Artifacts. Web search inside a session. Fast mode. The Advisor.

Different on Bedrock

Auto mode works only with Sonnet 5, Opus 4.7 and later, or Fable 5, and sessions start in Manual mode rather than auto. For web lookups the WebFetch tool with a specific URL still works. For scheduling, /loop inside a session stands in for /schedule.

What that means for the plan

Build the always-on agent, the channel, the phone control, and anything that verifies in a browser on the mini at home. Build the delegation habits, the hooks, the skills, the CLAUDE.md discipline at work, because they transfer both ways. The prompt cache works on both, with the note that on Bedrock its availability varies by model.

Read more