Thanks to visit codestin.com
Credit goes to openflows.dev

- Get Started

The Coder / OpenFlows Split

If you remember only one idea from this documentation, let it be this founding asymmetry:

Coder governs where agents run. OpenFlows governs how they coordinate.

This single split explains the entire architecture, and it's worth understanding on its own.

Two providers, two jobs

OpenFlows composes two very different systems, and each does the thing it's best at - nothing else.

CoderOpenFlows
JobWhere work runsHow work is coordinated
ProvidesIdentity and login, disposable workspaces, model gateway, audit, network policyThe team, the plan gate, adversarial review, self-healing, merges
NeverDecides what the team should buildDecides where work executes

Neither duplicates the other's core job.

Why the split makes the system secure

Most of OpenFlows' security comes directly from this division:

PropertyHow the split delivers it
AI calls happen only in the control planeThrough a central gateway, which means worker workspaces never hold an LLM key.
Git identity is a scoped OAuth linkNo shared raw token ever sits in a workspace.
Worker workspaces are thin and disposableNo framework, no keys, tight network - because the intelligence they need lives centrally, not inside them.

If OpenFlows tried to do Coder's job, it would have to put intelligence and credentials into every workspace, and the safety picture would collapse.

Why the split makes the system maintainable

ReasonWhat it means
Each side upgrades independentlyCoder can own and improve the agent loop; OpenFlows owns coordination.
One integration to keepEvery role is driven over one consistent interface, and OpenFlows stays a layer on top.
The pieces stay small and focusedCoder doesn’t think about workflow; OpenFlows doesn’t think about machines.

Where the trusted boundary sits

There is one long-lived, trusted component - the orchestrator workspace - which legitimately holds the credentials and runs the control loop. Everything else is ephemeral and untrusted. The split tells you exactly which box to trust and why, which is the mental model you need before reading the architecture material.

One trusted place, everything else disposable

Trust is concentrated rather than spread. Instead of sprinkling trust across many long-lived pieces, OpenFlows keeps it in a single auditable place and keeps the rest ephemeral.