Per-session isolation
Each task should run in its own sandboxed container with its own clone, so agents never share state and a misbehaving one has a small blast radius.
Run coding agents on your own server.
A self-hosted AI coding agent IDE lets you delegate real engineering tasks to an AI agent that runs on infrastructure you control — your code, your credentials, your network policy. Here's what the category is, what to look for, and how ShipIt fits it.
DEFINITION
A self-hosted AI coding agent is an agent that reads a task, writes and edits code, runs commands, and opens pull requests — running on a server or laptop you own rather than a vendor's cloud. The appeal is ownership: your repository lives and is worked on inside infrastructure you control, you set the egress rules, and you keep using the agent subscription you already have. Self-hosting the platform is not the same as self-hosting the model — the agent still talks to its provider unless you point it at a local one.
ShipIt is a self-hosted IDE built for exactly this. It runs Claude Code or Codex in isolated Docker containers on your VPS or laptop, gives each task a live app to test against, and surfaces the GitHub PR/CI/deploy loop and your issues inline — open source under Apache 2.0.
FIG. 1 — WHAT TO LOOK FOR
What a good self-hosted agent IDE provides
The agent is the commodity. The value is the loop around it — isolation, a real app to test against, the ship cycle, and security you can reason about.
A live app to test against
The agent should build and run a real instance of your app — not just edit files — and be able to start, stop, and tail the services it needs, so it can verify changes before you review.
Bring your own agent
Agent-agnostic by design: use the Claude Code or Codex subscription you already pay for, not a locked-in proprietary model.
The GitHub loop, inline
PRs, CI status, review threads, and deploy status surfaced where you work, so the ship loop doesn't scatter across tabs.
Strong security defaults
Container isolation, a default-deny outbound firewall per agent, brokered credentials rather than tokens sitting at rest, commit-time secret scanning, and guardrails around destructive git.
Own your infrastructure
Runs on a VPS or laptop you control, with no vendor platform in between holding your repository.
Why self-host coding agents at all?
Coding agents like Claude Code and Codex are already capable. The question is where they run. With a hosted product, your repository is checked out inside a third party's platform, you inherit their network and retention policies, and you're often tied to one vendor's model. Self-hosting flips that: the agent executes on infrastructure you own, so your source sits and is worked on where you control it, you decide what outbound traffic is permitted, and you can bring whichever agent you already trust and pay for. What it doesn't change is the model: unless you run a local one, the agent still calls its provider. On an always-on VPS, long-running and parallel tasks also keep going when your laptop is closed.
How ShipIt approaches it
ShipIt is a chat-driven IDE that orchestrates agent CLIs through your real engineering loop, with its control plane, workspaces, and previews entirely on your hardware. Each task is a session in its own isolated Docker container, with a full repo clone and a live instance of your app via Docker Compose to build and test against — services the agent can start, stop, and tail itself, with a JDK, the Android SDK, and Gradle in the image so Android builds and a live emulator work the same way. The GitHub PR lifecycle — description, CI checks, review threads, deploy status, merge state — renders inline in chat, next to your Linear and GitHub issues, down to a one-click sync when a branch falls behind its base and sessions that wake on their own PR merge. Security is treated as a first-class concern: agents run container-isolated as an unprivileged user with no Docker socket, outbound traffic is default-deny behind an allowlist, git credentials are brokered on demand, the post-turn auto-commit blocks commits carrying a recognized secret, and destructive git on already-merged work is routed through a brokered reset that leaves a record. You can run it on local Docker in minutes or provision an Ubuntu VPS with one command, optionally behind Cloudflare Zero Trust or Tailscale.
Honest limits
Self-hosting isn't free of trade-offs. You run the infrastructure — Docker on a machine you own — and each active session uses real resources (plan for 8 GB RAM minimum, 16 GB recommended on a VPS). ShipIt is in an early public release, it's designed as a single-tenant tool today, and its full review-and-ship loop depends on GitHub. If you'd rather run nothing and accept a vendor cloud, a hosted product will be lower-effort. The category exists for teams who specifically want ownership and control over convenience.
FIG. 2 — FAQ
Frequently asked questions
What is a self-hosted AI coding agent?
It's an AI coding agent — software that can read a task, write and edit code, run commands, and open pull requests — that runs on infrastructure you control rather than a vendor's platform. You host the orchestration, workspaces, and execution on a server or laptop you own, so no third party operates the system that holds your repository. ShipIt is one example: it runs Claude Code or Codex in isolated Docker containers that you self-host.
Why run coding agents on your own server instead of a cloud product?
Control and data residency. When an agent runs on your infrastructure, your repository is checked out and worked on there rather than inside a vendor's platform, you decide the network egress policy, and no third party holds your source at rest. It also lets you keep using the agent subscription you already pay for, and run on always-on hardware (a VPS) so long-running and parallel tasks continue even when your laptop is closed.
What should a self-hosted agent IDE provide beyond the agent itself?
The agent is the easy part — the surrounding loop is the hard part. Look for per-task isolation (a container per session), a live instance of your app to test against, the GitHub PR/CI/deploy loop surfaced where you work, issue tracking, parallel sessions you can review independently, and strong security defaults like container isolation, an outbound firewall, and brokered credentials rather than tokens sitting at rest in the container. ShipIt is built around exactly this loop.
Can I self-host an AI coding agent on a laptop, or do I need a VPS?
Both work. ShipIt installs with one command on local Docker (laptop) or on an Ubuntu VPS for an always-on setup. The VPS path is best when you want agents, previews, and CI follow-up to keep running when your laptop is closed; local Docker is the fastest way to try it.
FIG. 3 — COMPARE
How ShipIt compares
Evaluating against specific tools? These are factual, balanced breakdowns — including where ShipIt is the weaker pick.
NEXT STEP
Self-host your coding agents today.
One command installs ShipIt on your laptop or a VPS. Bring Claude Code or Codex and keep everything on infrastructure you own.