Run department agents with guarded access and memory

This setup lets Claude Code and Codex work as ongoing department agents with approvals, schedules, and shared memory.

Done with Claude CodeCodex· Advanced

What they did
OtoDock places Claude Code and Codex inside per-session kernel sandboxes, and it keeps network access shut off unless a folder or service is opened through MCP tools. A dashboard shows each action as it happens and pauses for permission when something risky comes up. The agents can be launched from webhooks or timed jobs, retain editable memory, hold meetings with one another, and create office documents and PDFs.
How it went
The README does not list any performance totals. The creator says agents on the system built and released much of OtoDock itself. Twilio phone calling is mentioned as a later shipping feature, not something users can use yet.
Tips
Keep a human approval step for anything that could not be undone easily, especially sending, removing, submitting, or paying. Setup requires a Docker Compose configuration and a Postgres password, and each user connects their own Claude or ChatGPT subscription, API credentials, or local model.

Try it yourself with Claude Code

You are my persistent [department] agent. Work from [folder or service] in a sandbox, using [Claude Code/Codex], and only access tools I explicitly allow through MCP. Keep a running memory of decisions, watch for [schedule/webhook] triggers, and coordinate with any other agents when useful. Show each step in the dashboard, then stop and ask me to approve before you do anything irreversible, including sending, deleting, submitting, or making payments. Your job is to help with [goal] and produce the needed deliverables in [document formats], while explaining any important tradeoffs first.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.