Three Orchestration Patterns, Actually Running
Real, compiled, tested LangGraph graphs for three of the five patterns in Agent Orchestration Patterns — including a genuine pause-and-resume approval gate, not a polling loop pretending to be one. Free to clone and run.
Same negation-aware escalation ladder as Quorel and RAG Reference Pipeline, the third implementation of the same regression-tested logic. Zero API key required to run the tests.
Some parts of a coordination pattern don't survive being drawn.
A diagram can show a supervisor routing to a worker. It can't show what actually happens to a graph's execution state while it's paused waiting on a human. This pack exists for the parts that only show up when the code runs.
Real conditional routing via a shared decision interface — not a diagram's arrows, an actual add_conditional_edges call deciding where execution goes next.
Genuine pause-and-resume via LangGraph's interrupt() and Command(resume=...) — execution actually suspends mid-graph and picks back up exactly where it stopped.
The same negation-aware algorithm as Quorel, this time as a real conditional edge between the deterministic check and the judge node.
Start free. Pay when it's carrying real traffic.
Same model as Quorel and RAG Reference Pipeline. The free tier is the actual patterns, not a crippled trial. Clone it, run it against your own use case, decide if it fits before you owe anything.
Full source for all three patterns, the shared AgentBrain stub, and 19 tests — everything you need to see how these actually run before you commit to anything.
- Supervisor/Worker with real conditional routing
- Human-in-the-Loop with real interrupt/resume
- Deterministic-First Escalation, Quorel's algorithm
- 19 passing tests, zero API key required
- Not licensed for production traffic
Everything you need to put these patterns in front of real traffic, plus guidance for wiring in a real model and a persistent checkpointer.
- Everything in Clone & Run
- License to run in production
- Walkthrough: wiring a real AgentBrain implementation
- Walkthrough: persistent checkpointer for pattern 4
- Email support for integration questions
The gaps that are genuinely specific to your environment, closed by hand instead of self-serve.
- Real model wired to your AgentBrain implementation
- Postgres-backed checkpointer for pattern 4
- Patterns 2 or 3 built for your specific workflow
- Combining patterns into one nested graph for your system
- Architecture review of your broader agentic system
clone & run proves the patterns work for your use case → production gets you shipped fast with real infrastructure → built for your stack closes what's genuinely yours to solve, not generic to productize.
If your multi-agent system is failing at a seam nobody explicitly designed —
the pattern probably exists already. The question is whether it was chosen on purpose, or assembled by accident.