Use cases
OpenFlows is a fit whenever you want AI to do engineering work inside your own infrastructure, under your own control, rather than handing code off to a hosted tool you cannot audit. Below are the situations where it shines.
An enterprise: governed, auditable AI delivery
The need. You want the speed of AI-generated features, but you must keep model keys private, keep a complete audit trail, enforce security policy, and stay within your own network.
How OpenFlows helps.
- The AI model runs behind your own model gateway, inside your own Coder control plane. Worker workspaces contain no model keys — there is no key sitting in a sandbox for an agent to exfiltrate.
- Every step — plan approval, review verdict, merge — is a typed, recorded event, so you can audit exactly what happened and why.
- Network policy is tight: agent workspaces can only reach the control plane, GitHub, and the coordination store. Everything else is denied.
- One Coder server can serve many teams at once, each isolated from the others.
A developer workflow: more throughput, same standards
The need. You have a backlog of issues and want AI to handle the mechanics — planning, implementation, testing, review — while you own the architecture and the final merge decision.
How OpenFlows helps.
- You stay in the loop only at security decisions, ambiguous specs, and major calls. Everything else runs autonomously.
- The planning gate means code is never written against an unagreed plan.
- The adversarial review means you don't have to re-litigate every pull request; the system already stressed it.
A team racing to ship: parallel, self-healing work
The need. Multiple issues flowing at once, with machines that crash and processes that stall.
How OpenFlows helps.
- OpenFlows runs a fleet of worker workspaces in parallel and assigns work to whichever is free.
- If a workspace crashes or goes silent, the system detects and recovers it automatically, with bounded retries.
- Completed pull requests are watched through CI and merged when green.
A solo developer: a personal AI team for your backlog
The need. You work alone, and you have a backlog of issues and features but only so many hours in a day. You want an AI team that does the grunt work and returns reviewed, mergeable changes.
How OpenFlows helps.
- Open an issue and let the team plan, build, test, and adversarially review it while you are away — you come back to a reviewed pull request, not a pile of work.
- You stay in control: you own the architecture and the final merge decision, and the team only taps you for the calls that genuinely need a person.
- It runs on top of your own self-hosted Coder, so even as a solo operator your private code never leaves your infrastructure and worker workspaces hold no keys.
A stakeholder: transparency from intent to shipped
The need. You want to see progress from "we want this feature" to "it's deployed," without digging through code.
How OpenFlows helps.
- Every issue becomes a ticket with a visible lifecycle: planned → built → reviewed → merged.
- Decisions that need a human are surfaced with notifications and direct links.
- A web control panel shows the live state of the whole fleet at a glance.
Extending the workspace: skills, models, and tools
The need. Different features need different capabilities — a new linting skill, a model better suited to a task, an MCP server for a private tool.
How OpenFlows helps. OpenFlows is plug-and-play. You can add a skill, register an MCP server, or enable a new model as configuration only — no engineering work required.
Enterprise and solo, on the same tool
When might it not be the right fit?
OpenFlows assumes you have a self-hosted Coder environment and a GitHub repository you control. It is designed for governed, in-your-network delivery. If you have no existing Coder infrastructure and want a fully hosted, zero-infrastructure AI coding assistant, that's a different trade-off.