{"slug": "otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox", "title": "OtoDock 1.6.0 lets you self-host Claude Code and Codex agents in a sandbox", "summary": "OtoDock released version 1.6.0, a Fair Source self-hosted platform that wraps Claude Code and Codex CLI into a multi-tenant app, running each agent as a persistent process inside a bubblewrap sandbox with pasta network isolation and a single outbound WebSocket. The release is free for up to 5 users, requires users to supply their own Anthropic or OpenAI API keys, and its code is public on GitHub; the company said a large portion of version 1.6.0 was written by the agents themselves using Claude Code.", "body_md": "# OtoDock 1.6.0 lets you self-host Claude Code and Codex agents in a sandbox\n\nSelf-hosting an \"AI company OS\" is actually viable if you use bubblewrap for kernel sandboxing and pasta for network isolation. I've been using OtoDock to run persistent agents that act as department heads, and the 1.6.0 update just landed. It basically wraps [Claude Code](/en/tags/claude%20code/) and Codex CLI into a multi-tenant app, meaning you can have different agents with their own workspaces and memory, all running on your own hardware via your own API keys or local models.\n\n## How the agent isolation works\n\nThe biggest risk with giving an agent CLI access is letting it wreck your host machine. OtoDock handles this by running each agent as a persistent process inside a bubblewrap sandbox. This keeps the filesystem isolated and uses pasta to manage network traffic without needing to open a bunch of inbound ports or mess with a complex VPN.\n\nIf you need the agent to interact with a remote computer, it uses a single outbound WebSocket. This is a huge win for security because you don't have to poke holes in your firewall for every single agent you deploy.\n\n## What these agents can actually do\n\nMost \"agent\" wrappers are just chat interfaces. These are persistent processes with actual tools. During my setup, I found a few specific integrations that aren't common in basic LLM wrappers:\n\n- **Document Handling:** It uses Collabora to let agents edit and preview Excel, Word, and PPT files directly in the chat.\n- **Communications:** Integration with Twilio and Asterisk allows agents to actually place and answer phone calls.\n- **Deployment:** It's a single install script and a docker-compose file. No signup process is required to get it running locally.\n\n## Getting it running and the cost\n\nThe license is Fair Source, and it's free for up to 5 users. Since you provide your own Anthropic or OpenAI keys, your only cost is the token spend and the electricity for your server.\n\nTo get it started, you'll need Docker installed. The setup is basically:\n\n```\n# Run the installation script provided in the repo\ncurl -sSL https://otodock.io/install.sh | bash\n```\n\nThen you launch the stack via docker-compose. If you run into issues with the sandbox, check that your kernel supports bubblewrap, as that's where most \"it won't start\" errors happen on older Linux distros.\n\n## When to avoid this setup\n\nDon't bother with this if you only need a simple chatbot or if you're uncomfortable managing a Docker environment. If you don't have a dedicated server or a high-performance VPS, the overhead of running multiple persistent agent processes might lag your system. Also, if you're looking for a purely SaaS experience where you don't manage keys, this isn't for you.\n\nThe code is all public on GitHub, and interestingly, a huge portion of version 1.6.0 was actually written by the agents themselves using [Claude](/en/tags/claude/) Code.\n\n```\nhttps://github.com/OtoDock/oto-dock\n```\n\n[Next Should AI companions be banned for under 18s to prevent emotional dependence? →](/en/news/9143/)\n\n## All Replies （5）\n\nCurious if you're using gRPC or WebSockets for the routing? The mobile Firefox bug sounds like a CSS grid issue.\n\nI'm so confused. Why would anyone use Blender for that? Seems like a total waste of render time...\n\nI'm curious if the latency is actually lower. Does this play nice with Terraform or is it a manual nightmare?\n\nI'm dying to try this tonight. Does it actually solve the memory leak in v2.4?\n\nCurious if this works with GPT-4o. I'm wondering if using 3 separate validators would actually stop the hallucination loop?", "url": "https://wpnews.pro/news/otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox", "canonical_source": "https://promptcube3.com/en/news/9170/", "published_at": "2026-09-10 18:38:29+00:00", "updated_at": "2026-09-10 19:07:40.128001+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-products", "developer-tools", "ai-safety"], "entities": ["OtoDock", "Claude Code", "Codex CLI", "Anthropic", "OpenAI", "bubblewrap", "pasta", "GitHub"], "alternates": {"html": "https://wpnews.pro/news/otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox", "markdown": "https://wpnews.pro/news/otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox.md", "text": "https://wpnews.pro/news/otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox.txt", "jsonld": "https://wpnews.pro/news/otodock-1-6-0-lets-you-self-host-claude-code-and-codex-agents-in-a-sandbox.jsonld"}}