OpenFlows Documentation
OpenFlows Manager governs users, runtimes, and workspace operations while Coder runs the agent workspaces. Read these docs to install, operate, extend, and contribute to the system.
Recommended path
Read the system in order
Start with the product, learn the component boundaries, then install and operate it. Each page links to the next.
- 01What is OpenFlowsThe product and the problem it solves.
- 02How OpenFlows worksFollow an issue through the agent team.
- 03Components and interactionsSee the system boundaries and who talks to whom.
- 04Manager runtimeUnderstand registration and workspace control.
- 05Credentials and trustLearn which credential belongs to each boundary.
- 06Hooks and A2ASee event delivery and worker verification.
- 07InstallationPrepare and configure a deployment.
- 08First issue walkthroughRun a complete workflow.
- 09Operations and monitoringKeep the system healthy and recover failures.
Browse by topic
Introduction
New to OpenFlows? Start here to understand what it is and when it fits.
What is OpenFlows
The core idea, whom it is for, and a quick tour of the team.
How OpenFlows works
The end-to-end flow from issue to merged pull request.
Use cases
The situations where OpenFlows shines.
How OpenFlows compares
OpenFlows vs. chat assistants, hosted agents, bare orchestrators, and Coder alone.
Get Started
Install OpenFlows, obtain tokens, and run your first autonomous issue-to-merge cycle.
Quick Start
The fast path from clone to a working OpenFlows stack.
Installation
What you need before you start and how the environment is put together.
First Issue Walkthrough
What happens when you create a GitHub issue and the agents take over.
Token Guide
User sign-in, runtime sessions, and brokered GitHub access.
Lifecycle Steps
The phases every ticket travels through, from planning to merge.
Completion & Merge
How work is reviewed, merged, and torn down.
Core Concepts
The ideas that make OpenFlows work: architecture-first orchestration, the agent team, and the Coder integration model.
Architecture-First Development
Why the engineering discipline is the product.
The Agent Team
NEXUS, FORGE, SENTINEL, VESSEL, and LORE - roles, boundaries, and handoffs.
The Planning Gate
Why agents must plan before they code.
The Coordination Store
The single record of truth that keeps every agent consistent.
Workspaces
Where work actually runs - and why it is safe.
Tenants
How one server serves many isolated teams.
Human Escalation
When and how a person gets pulled in.
The Agent Registry
How the team is configured without code.
The Coder / OpenFlows Split
The founding idea: where work runs vs. how it is coordinated.
Architecture
Manager-mediated runtime architecture, credential boundaries, agent coordination, recovery, and deployment - in plain language.
Components and Interactions
A component map of Manager, Coder, Nexus, workers, storage, hooks, and GitHub.
Manager Runtime
How workspaces register, receive runtime identity, and request lifecycle operations.
Credentials and Trust
Where platform secrets stay, what a runtime receives, and how GitHub access is brokered.
Hooks and A2A
How Coder lifecycle events, hooks, Nexus, and worker evidence flow together.
Big Picture
The control plane and the work plane, and how they fit together.
The Controller
The brain and its orchestration loop.
The Workspace Fleet
Ephemeral worker spaces and the long-lived NEXUS workspace.
How the Pieces Talk
The three channels that keep the team in sync.
State Machine & Recovery
NEXUS reconcile, failure detection, and self-healing.
Decision & Routing
How work is dispatched and steered between roles.
Security by Design
Why the architecture is safe by construction.
How It's Deployed
The reference deployment, stood up for many teams.
Guides
Practical instructions for configuring, extending, and operating OpenFlows in production.
Agent Setup
Configure roles, models, skills, and tools from the control panel.
Model Routing
Choose which model each role uses.
Multi-Tenancy
Running multiple teams on a single Coder deployment.
Extending: Skills
Teach a role something new, configuration only.
Extending: MCP
Connect external tools and data.
Extending: Models
Add a new model to the platform catalog.
Operations & Monitoring
Status, steering, notifications, and recovery runbooks.
Troubleshooting
Common failures, diagnostics, and fixes.
Reference
Quick lookup: roles, glossaries, lifecycle states, configuration, and the control surface.
Roles Reference
The five team roles and their responsibilities.
Glossary
Every term defined in one line.
Lifecycle States
Every state a ticket passes through, and the enforced rules.
Configuration Reference
The knobs you can set, grouped by concern.
Limits & Defaults
The meaningful numbers to know.
Command Surface
The programmatic way to drive the same operations as the panel.
Notifications Reference
Channels, what they carry, and how they behave.
Contributing
How to get involved and how the project is governed.
Resources
Further reading and where to get help.