Northflank

Filed by the Vibe Code Weekly desk · Updated Nov 12, 2026

The story

Northflank covers what happens after code is written: it is where agent-written code, and the agents themselves, execute. The company calls it a runtime for autonomous software, spanning short-lived sandboxes through to live production, hosted on Northflank's own cloud or a customer's.

In the AI-coding stack, Northflank is infrastructure rather than an editor or app generator. It is built for four kinds of users. Engineers maintain APIs, databases and critical services. Vibe coders publish internal tools. Coding agents run tasks and open pull requests. AI products execute code submitted by their own users. All four feed into one control plane that builds, releases, scales, secures, observes and recovers workloads on Kubernetes.

The agent-facing pieces are the newest part. Every workload gets its own microVM, so isolation is the default. Through the API, a sandbox can be created from an image such as python:3.12 and a command executed inside it. Cloud agents let harnesses such as Codex, Claude Code, OpenCode and Pi run remotely with persistent storage and real compute. AI-built internal apps, for example a sales forecast or a vendor portal, can sit behind SSO on the same platform as production systems.

The rest is a conventional platform-as-a-service. Services build from Git or any image, using a Dockerfile or buildpacks. Managed Postgres, MySQL, MongoDB and Redis are available. Cron and on-demand jobs cover batch work and migrations, and H100, A100 and L4 GPUs serve inference and training. Workloads can run on Northflank Cloud across 17+ regions, inside AWS, Google Cloud, Azure, Oracle, CoreWeave or Nebius accounts, or on bare metal, OpenShift and VMware, including air-gapped sites.

Coding agents are now opening pull requests and running tasks on their own, so the code they produce needs a safe place to execute. Northflank positions itself as that runtime, from short-lived sandboxes through to live production, on its own cloud or a customer's.

The good

MicroVM isolation applies to every workload, not only sandboxes. Agents, internal apps and production services share one control plane. API access lets agents create sandboxes and run commands programmatically. Deployment targets range from Northflank Cloud to BYOC and on-premises hardware. Release tooling includes previews, pipelines and rollbacks alongside audit logs and backups.

The catch

It hosts code but does not generate it, so a separate coding tool or agent is still required. The number of cloud and hardware targets means more choices for a small team to make.

Who should read on

Teams giving coding agents remote compute and persistent storage. AI products that must execute untrusted user code in isolation. Companies moving vibe-coded internal tools onto governed hosting with SSO. Organisations wanting a managed platform inside their own cloud account or data centre.

Who can skip it

People who want a tool that writes the application from a prompt. Software that cannot be packaged into a container image.

Reader mail