Quick Start Guide
Getting Started
OpenFlows is an autonomous AI software team that turns GitHub issues into reviewed, production-ready pull requests, running entirely inside your own self-hosted Coder environment. It stands up a small team of specialized agents that plan, write, test, and adversarially review code on your behalf. You stay in control of the important decisions; the team handles the execution.
The one-line idea
1Who it is for
OpenFlows is an enterprise-grade orchestration layer — self-hosted, governed, auditable, and multi-tenant — that is just as at home for a solo developer who wants a personal AI team. Anyone who wants AI to execute end-to-end with guardrails inside their own environment fits here:
- Enterprises & teams that need governed, auditable AI delivery inside their own network, with strict isolation and no model keys in worker workspaces.
- Developers who want the mechanics of planning, coding, testing, and review handled while they stay in charge of the important decisions.
- Solo developers who want a personal AI team to work through their backlog and return reviewed pull requests, without babysitting every change.
2Prerequisites
Coder governs where agents run; OpenFlows governs how they coordinate
- Coder decides where work runs. It provides governed, isolated, disposable workspaces and a central model gateway where every AI call happens. Worker workspaces hold no model keys, no raw repository credentials, and no secrets.
- OpenFlows decides how the work is coordinated. It provides the team, the planning gate, adversarial review, self-healing, and merges.
3Install & Set Up
Setting up OpenFlows means standing up the control plane (Coder, the coordination store, and a database), running the installer to bootstrap the environment, adding a tenant for your repository, and linking GitHub OAuth. From there you can open an issue and watch the team turn it into a merged pull request.
Step-by-step guidance lives on the Installation page and the First Issue Walkthrough.
4Track Your First Issue
Once the environment is up and your tenant is configured, create a clear GitHub issue in the bound repository. OpenFlows does the rest:
- NEXUS picks the issue up and turns it into a ticket.
- FORGE writes a plan, then halts.
- SENTINEL reviews the plan and approves it.
- FORGE implements, tests, and opens a pull request.
- SENTINEL reviews the pull request adversarially.
- VESSEL watches CI, resolves conflicts, and merges.
- LORE documents the change (where enabled).
- The workspace is torn down after merge - no lingering state.
If at any point OpenFlows needs a human call - a security concern, an ambiguous spec, or an unresolvable merge - you are notified and the team waits for your decision. See the First Issue Walkthrough and the Lifecycle pages for the full picture.
5What OpenFlows is not
- Not a chat assistant. It is not a tool you prompt and babysit; it is a team that plans, reviews, and merges under its own guardrails. You decide, it executes.
- Not a hosted cloud coder. It runs in your own Coder control plane. Code, prompts, and outputs stay in your environment, and worker workspaces hold no keys by design.
- Not a bare orchestrator you maintain. The planning gate, adversarial review, recovery, and human escalation are already built - you do not write the coordination logic.
- Not a replacement for Coder. It is a layer on top. It needs your self-hosted Coder environment to run, and it favors governance over raw speed.
6Next Steps
Digging deeper
- Coder Integration - how OpenFlows and Coder split the work
- Architecture-First - why OpenFlows plans before coding
- The Agent Team - who NEXUS, FORGE, SENTINEL, VESSEL, and LORE are
- Security Model - how secrets stay out of worker workspaces
- Operations - managing the fleet at runtime
- Multi-Tenancy - one Coder server, many teams