Claude Code self hosted environments let your Claude Code cloud sessions run on infrastructure your own organisation operates, instead of on Anthropic-hosted machines — which means a session started from the web, desktop, mobile or a scheduled routine executes inside your network, right next to your internal services, toolchains and security controls. For anyone automating real workflows against private systems, that is the missing piece: the agent can finally reach the databases, registries and internal tools it needs without you exposing any of them to the public internet. Everything here is drawn from Anthropic's official Claude Code documentation and announcement; where a detail matters, it is attributed to the source.
📺 Watch: NEW Claude Update is CRAZY GOOD!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
What Claude Code self hosted environments actually are
According to Anthropic's official documentation at code.claude.com, self-hosted environments are in public beta on the Team and Enterprise plans and are off by default. The core idea is a change of where the work happens. Normally a Claude Code cloud session runs on Anthropic's infrastructure. With a self-hosted environment, that same session executes on compute your organisation controls — your own servers, your own cloud account — so the agent operates from inside your perimeter rather than outside it.
Anthropic first opened this beta to Team and Enterprise customers in mid-August 2026, per its announcement, "Self-hosted environments for Claude Code." The framing in the documentation is deliberately unglamorous: this is for teams whose network, tooling or compliance requirements call for keeping agent execution on infrastructure they control, and those teams should plan to staff engineering to own the setup and ongoing maintenance. In other words, it is a capability, not a convenience toggle.
Why this matters for automation
Most automation projects die at the same wall: the agent cannot see the thing it needs to act on. Your database sits on a private network. Your package registry is internal. Your staging services are firewalled off from the public internet, exactly as they should be. A cloud agent running on someone else's infrastructure is on the wrong side of that wall.
Self-hosted environments move the agent to the right side of it. The documentation states that sessions run inside your network and can reach internal services, databases and registries without exposing them to the public internet. That single property unlocks a category of automation that was previously impractical: an agent that can query your real internal data, run against your real internal tools, and do it under your existing network and security controls rather than around them.
Custom tooling comes along for the ride
The second benefit the documentation highlights is custom tooling. Organisations can pre-install compilers, software development kits and internal command-line tools so they are available during sessions. If your build needs a specific toolchain, or your workflow depends on an in-house CLI, you no longer have to reconstruct that world inside a generic sandbox — the agent runs where those tools already live. For automation that touches your actual build and deployment pipeline, that is the difference between a demo and something you can rely on.
If you want to build agent workflows that plug straight into your real systems and actually save you hours every week, check out the AI Profit Boardroom, where members get the full Agent OS zip, ready-made automations and weekly coaching to put them to work → join the AI Profit Boardroom. Prefer a tailored plan for your own setup first? You can book a free SEO strategy session and we will map out where automation will pay off fastest.
What Claude Code self hosted environments do not change
This is the part worth reading twice, because the name can mislead. Self-hosting the execution environment does not mean self-hosting the model. The documentation is explicit: prompts, model responses, tool results and session transcripts still travel to Anthropic for inference, and Anthropic retains the transcript so work can continue across the web, desktop, mobile, terminal and scheduled-session interfaces.
So the correct mental model is this. The compute that runs your tools, touches your files and reaches your internal services lives with you. The intelligence that decides what to do still comes from Anthropic's models over the network. If your reason for wanting self-hosting is to keep every byte of prompt text on your own metal, this feature on its own does not deliver that, and the documentation says so plainly. If your reason is to let the agent reach private infrastructure and use internal tooling under your own controls, it is exactly the right tool.
There is a practical upside to that design worth calling out for automation builders. Because Anthropic retains the session transcript, the documentation notes that work can continue across the web, desktop, mobile, terminal and scheduled-session interfaces. For an automation workflow that means a job kicked off by a scheduled routine and a follow-up you make later from your desktop share the same thread of context — the agent does not start from zero each time. When you are chaining unattended runs against internal systems, that continuity is what stops each execution from being an island, and it is a direct consequence of where the transcript lives rather than where the compute runs.
Who should turn it on
The honest answer is: teams with a specific need and the engineering to support it. If you are a solo operator automating public-facing tasks, the standard cloud environment is simpler and you will not miss anything. The self-hosted path earns its keep when you have private data the agent must reach, internal tooling it must use, or compliance rules that require agent execution to stay on infrastructure you own. Because it is off by default and asks you to staff its upkeep, it is designed for teams that will get real value from that control — not as a box everyone should tick.
| Question | Standard cloud environment | Self-hosted environment |
|---|---|---|
| Where sessions run | Anthropic infrastructure | Your own compute |
| Reach internal services and databases | Not without exposing them | Yes, inside your network |
| Pre-installed internal tooling | No | Yes, you provision it |
| Where inference happens | Anthropic | Anthropic (unchanged) |
| Setup and maintenance burden | None | You own it |
How it fits an automation stack
Self-hosted environments are an infrastructure decision, but they pay off through the workflows you build on top. Once the agent can reach your internal systems, the automations you design need a consistent structure so they behave the same way every time, which is exactly what a setup like Agent OS gives you. If you are weighing which model to run for which job across those workflows, the Goldie Bench comparison covers how the leading models perform in hands-on tests, so you can match capability to task.
It also sits naturally alongside the ways people already run Claude Code in controlled conditions. If you have been keeping workloads off the metered path by running Claude Code locally, self-hosted environments extend that same instinct to team-scale cloud sessions. And because the agent now operates inside your perimeter, the permission and access controls you already care about matter more than ever — the guides on restricting Claude Code access and tuning Claude Code permission prompts are worth reading before you point an agent at anything sensitive.
Getting the most out of it
Two practical habits make self-hosted environments safer and more useful. First, treat access as something you grant deliberately rather than by default: give the agent the internal services it needs for the task in front of it, not a blanket key to everything. Second, connect it to the internal tools you actually want automated — an MCP server that exposes a private system to the agent is a clean way to do this, and the walkthrough on setting up an MCP server shows the pattern. Together these turn a raw capability into a workflow you can trust to run unattended.
If cost is what is holding you back from experimenting, remember you can prototype the workflow logic cheaply first — the guide to running Claude Code for free covers the no-cost paths for building and testing before you commit engineering time to a self-hosted rollout.
The bottom line
Claude Code self hosted environments are a targeted, powerful feature: they put agent execution on infrastructure you control so your automations can finally reach the private services, data and tooling that real businesses run on, all under your existing security controls. They do not move inference off Anthropic's models, and they ask you to own their setup and upkeep — so they are built for teams with a genuine need, not for everyone. If that describes you, the payoff is an agent that works where your systems actually live, which is where the most valuable automation has always been waiting to happen.
Want help turning this into automations that reach your real systems and save you serious time? Inside the AI Profit Boardroom you get the Agent OS, done-for-you workflows and coaching to build it properly → get inside the AI Profit Boardroom. And if you would like a personalised roadmap for your own stack, book a free SEO strategy session and we will plan it together.
