Eclipse Enclave 1.0: Your Coding Agent, Sandboxed
Eclipse Enclave 1.0, announced November 19, 2026, runs each coding agent in its own isolated container, with only the files, network hosts and secrets you choose. Free, open source, MIT licensed.
AI Stella and Liam are AI hosts with AI-generated voices. About our AI hosts
Key takeaways
- The Eclipse Foundation announced Eclipse Enclave 1.0 on November 19, 2026. It is free, open source, and MIT licensed, and the project registry dates version 1.0.0 to October 31, 2026.
- Each agent runs in its own isolated container with only the files, network hosts and secrets you choose. Everything else on your machine, like SSH keys and other projects, stays out of reach.
- Six agents are built in: Claude Code, Codex, OpenCode, GitHub Copilot CLI, Pi, and Eclipse Theia, which can attach to a running session. Amp and Antigravity ship separately in the enclave-extensions repository, and tooling can also be pulled from git repos.
- Enclave runs on Linux and macOS including arm64, with deb and rpm packages and a Windows launcher via WSL2. Backends are Docker in rootful mode with buildx and Podman with rootless support, plus an experimental QEMU microVM backend.
- The honest limits: the microVM backend currently operates without the Docker gateway network restrictions, HomeShell is a preview with no stable release, a sandbox limits damage but does not judge code quality, and database credentials should point at a development schema only.
Show notes
The Eclipse Foundation announced Eclipse Enclave 1.0 on November 19, 2026. Enclave is a free, open source, MIT licensed sandbox that runs each coding agent in its own isolated container, with only the files, network hosts and secrets the developer chooses. Six agents are built in: Claude Code, Codex, OpenCode, GitHub Copilot CLI, Pi, and Eclipse Theia, which can attach to a running session. Amp and Antigravity ship separately in an enclave-extensions repository, and tooling can also be pulled from git repos. It runs on Linux and macOS including arm64, with deb and rpm packages, and a Windows launcher via WSL2. In this episode of Agentic Pod, Liam and Stella explain how the sandbox works, which backends it uses, and the caveats to know before you trust it.
What we cover:
- What shipped: Eclipse Enclave 1.0, announced November 19, 2026, free and open source under the MIT license, download at enclave.eclipse.dev
- The sandbox: each agent in its own isolated container, with only the files, network hosts and secrets you choose
- Six built-in agents: Claude Code, Codex, OpenCode, GitHub Copilot CLI, Pi, and Eclipse Theia attaching to a running session
- More tooling: Amp and Antigravity in the enclave-extensions repository, plus tooling pulled from git repos
- Platforms: Linux and macOS including arm64, deb and rpm packages, Windows launcher via WSL2
- Backends: Docker in rootful mode with buildx, Podman with rootless support, and an experimental QEMU microVM backend
- The caveats: the microVM backend currently operates without the Docker gateway network restrictions, HomeShell is a preview with no stable release, and the registry dates version 1.0.0 to October 31, 2026
- The limits: a sandbox limits damage but does not judge code quality, review everything, and point database credentials at a development schema only
Chapters
Questions answered in this episode
What is Eclipse Enclave?
A free, open source, MIT licensed sandbox from the Eclipse Foundation, announced November 19, 2026, that runs each coding agent in its own isolated container with only the files, network hosts and secrets you choose.
Which agents does Enclave run?
Six are built in: Claude Code, Codex, OpenCode, GitHub Copilot CLI, Pi, and Eclipse Theia. Amp and Antigravity ship separately in the enclave-extensions repository, and tooling can also be pulled from git repos.
What does Enclave cost?
Nothing. It is free and open source under the MIT license, and you run it on your own machine. Download it from enclave.eclipse.dev.
Does a sandbox make agent code safe to ship?
No. A sandbox limits the damage an agent can do while it works. It does not judge code quality, so review everything the agent writes, and point database credentials at a development schema only.
Links & resources
Full transcript
Intro
[excited] Hey hey hey, welcome to Agentic Pod! I'm Liam.
And I'm Stella! Liam, the Eclipse Foundation announced Enclave 1.0 on November 19th, and it puts your coding agent in a sandbox?
A real sandbox. Each agent runs in its own isolated container, with only the files, network hosts, and secrets you choose to give it.
So the agent can run free inside the box, and the box stays small?
Exactly that. By the end you will know how it works, which agents it runs, and the one caveat you must not skip.
What Enclave actually is
Okay, plain words. What exactly is Eclipse Enclave?
It is a free, open source sandbox from the Eclipse Foundation. MIT licensed, so anyone can use it and read the code.
MIT licensed means no fee, no lock in?
Right. You download it from enclave.eclipse.dev and run it on your own machine.
And the container part. Explain that like I have never run one.
A container is a sealed room for one program. The agent sees the project files you mount, and nothing else on your machine.
So my SSH keys and other projects stay out of reach?
Out of reach. You pick the files, the network hosts it may call, and the secrets it may use. Everything else stays outside the walls.
[laughs] A tiny apartment for my agent, with a lock I control. So which agents can move in?
Six agents built in, and how to try it
Six are built in. Claude Code, Codex, OpenCode, and the GitHub Copilot CLI.
That is four. Keep going.
Pi makes five. And Eclipse Theia makes six, and Theia can attach to a session that is already running.
Attach, meaning I can open a window into the live session and look around?
Exactly. And two more live next door. Amp and Antigravity ship separately, in a repository called enclave-extensions.
And if my tool is not on either list?
Enclave can also pull tooling straight from git repos, so you are not stuck with the built in set.
Okay, trying it. Step by step.
Download Enclave from enclave.eclipse.dev. Install it on Linux or macOS, including arm64 machines. On Linux you get deb and rpm packages too.
Windows?
Windows gets a launcher that runs it through WSL2. Then you start a session, pick your agent, and mount only the project it should touch.
Simple enough. But containers need an engine underneath, right? That is where it gets picky, I bet.
Backends, HomeShell, and the honest caveats
It does get picky. Enclave runs on two container backends. Docker works in rootful mode, with buildx.
Rootful, meaning Docker runs with full system rights?
Yes. Podman is the other backend, and there rootless works, so it runs without full system rights.
And I heard there is a third, experimental one.
The QEMU microVM backend. It runs the agent inside a tiny virtual machine. Here is the honest caveat. Right now it operates without the Docker gateway network restrictions.
So the microVM path does not yet filter network calls the way the Docker path does?
Correct. Treat it as experimental, and test it on work you can afford to redo. There is also HomeShell, a graphical companion.
HomeShell shows me my sessions?
It groups your sessions into projects, so parallel work stays tidy. But it is a preview. There is no stable release yet.
Preview means pretty, but not promised. Noted. One more date thing is bugging me though.
Dates, limits, and the honest verdict
The registry says version 1.0.0 landed on October 31st, but the announcement was November 19th. Which date counts?
Both are real. The project registry dates 1.0.0 to October 31st, 2026. The Eclipse Foundation announced it in public on November 19th, 2026.
So the code shipped first, and the party came later.
[laughs] A quiet ship, then a loud launch. Now the limit that matters most. A sandbox limits damage. It does not judge code quality.
Meaning bad code inside the box is still bad code.
Still bad code. Anything the agent writes still needs your review, same as always. And one practical rule. Point database credentials at a development schema only, never production.
So even inside the sandbox, give it the spare keys, not the house keys.
Perfect way to say it. My verdict? If you let agents run on real projects, this is the seatbelt. Free, MIT licensed, six agents built in, and you choose every file and secret.
My take? Try it this week on one small project. Mount one folder, allow two network hosts, and watch what the agent asks for. Would you let an agent run unattended inside a box like this, or do you still watch every step? Tell us in the comments. See you next time!