CodeHerderSearch⌘KRequest access →

A fresh, locked-down workspace for every agent.

Every stage of every task runs in a new, hardened container. It is built from a pinned base image plus the tools your team declares. Your code stays on your devices, and one dedicated proxy is the only way out.

Built in layers. You control the top ones.

CodeHerder maintains a hardened base. Your team adds what its projects need. Every part is pinned and verified, then cached, so each stage starts the same way every time.

Hardened base

Pinned, patched, scanned

The base image is pinned by digest, and its packages come from one frozen snapshot. Every download is checksum verified, and the image is scanned before any device receives it.

Your additions

Declared, not hand-built

Name the tools and packages your projects need. CodeHerder resolves exact versions and verifies them. Then it builds the result on your own device.

Ready first

No waiting at the start line

Environments build ahead and stay cached. When something changes, stages keep using the previous build until the new one is ready.

Make it yours without maintaining images.

Three settings cover almost every project. Set them once for a group of workspaces, then override them for one workflow or one stage.

groupworkspaceworkflowstagewhat the agent gets

tools

Exact versions of the CLIs you use

Pick from a large open-source tool registry. Language packages work too. Each version is pinned and checksum verified, and where the publisher signs a release, CodeHerder checks the signature. If a tool's download source changes, an admin confirms it first.

system

OS packages and custom install steps

Add Debian packages and a short Dockerfile fragment. CodeHerder builds them on top of the hardened base, on your device, behind a locked-down build network.

setup script

Setup for each checkout

Install dependencies or run codegen as each stage starts. Secrets come from encrypted workspace variables, never from the script.

ch · shape an environment
# find a tool and pin it
❯ ch env search terraform
❯ ch env tools add hashicorp/terraform 1.9.8
pinned hashicorp/terraform 1.9.8
# add an OS package and a setup step
❯ ch env system add libpq-dev
❯ ch env set --from-file env.json
# see exactly what one stage gets
❯ ch env show --effective --task <id> --stage code
❯ ch env status

Every change is versioned. Preview a change before you save it, and revert to any earlier version.

Your code never leaves your devices.

Images hold tools, never code. Repositories live in git on the machines you register. Each stage gets them mounted in. CodeHerder keeps the record of the work, not the work itself.

One way out, and you decide where it goes.

Each run gets its own private network with no direct route anywhere. Its own proxy is the only exit, and it checks every connection.

  • Internal addresses are always blocked. That covers cloud metadata endpoints and private networks, including loopback. The check uses the address a name actually resolves to.
  • Git credentials stay with the device. The proxy relays git and API calls to the device, which adds access for the repos the task may use.
  • Only stages that may write can push. Read-only stages can read the repo but cannot push to it.
  • New hosts need a yes. A blocked request becomes an approval request. An admin opens one exact host, and the running agent gets it at once.
  • Image builds have their own rules. A host opened for a build never opens for an agent, and the reverse.
  • Every approval is recorded. Requests and approvals sit in the audit trail.

The detail is in agent isolation and network access approvals. Reviewing us for your security team? Read the security review answers.

The same environment, wherever agents run.

Define your environment once. Every isolated device builds it the same way, from the same base, behind the same proxy.

Your hardware

Any machine you run with Docker, at the office or in your data centre.

AWS EC2

Dedicated, always-on hosts, launched in one click.

Lambda microVMs

Environments built ahead into the image, so stages start warm.

See every way to connect one on where agents run.

Questions people ask about agent environments

Can we bring our own Dockerfile?

Yes. Add a Dockerfile fragment on top of the CodeHerder base, or run a stage in your own container image. Most teams never need to: the three environment settings cover what they need.

How are tool versions controlled?

Every tool is an exact version. CodeHerder resolves it to verified checksums for each platform and runs the publisher's signature checks where they exist. If a tool's download source changes, an admin confirms it first.

Does CodeHerder store our code?

No. Repositories live on your devices, in git. CodeHerder stores metadata about the work, such as commit IDs and the paths each commit changed.

Can an agent take our git credentials?

On an isolated device the credential never enters the agent's container. The run proxy relays git traffic to the device, which adds access for the repos that task may use.

What stops an agent from calling any host it likes?

The agent has no direct route out. Internal addresses are always blocked, and on stages that can change code a new host needs an admin's approval.

Do package installs go around the proxy?

No. Package downloads go through the same proxy to a shared cache on the device. Environment builds run behind their own locked-down network.

Does the environment differ between our machines and the cloud?

No. Every isolated device builds the same environment from the same base, behind the same proxy, whether it runs on your hardware or in the cloud.

Give every agent a clean room to work in.

Declare your tools once. CodeHerder builds and verifies the rest, then keeps each agent isolated.

Try "pricing", "connect a device", or "who reviews the code"

↑↓ move · ↵ open · esc close