AI GraphQL Flow
Asynchronous GraphQL workflow engine prototype: a declarative GraphQL API (Yoga, serverless-first for Vercel) persists a workflow graph and enqueues steps for distributed workers, with a package that dynamically generates the schema and types.
- BPM
- 133
- GRADE
- B
- LEVEL
- 4
- RELEASES
- 0
- YEARS
- 2025–2025
A Graph That Runs Itself
In late August 2025 I had an idea: describe a workflow in GraphQL and let the API be the queue.
You post a graph, it gets saved, and its steps get queued up for distributed workers to pick up, so a mutation that would normally block for minutes returns right away and the work happens somewhere else. GraphQL Yoga on the server, serverless-first so it deploys to Vercel, and a package that generates the schema and the types from the workflow definition so you don't have to write both by hand. That last part was my favorite bit.
The "ai" in the name is because a chain of model calls is exactly this kind of workload: slow, expensive, and worth resuming instead of retrying. It was a sketch more than a finished runtime, but durable step execution kept showing up in what I built after, and Hypery started around the same time as an AI gateway with auth and billing.
led to
- Nihmbus2026async workflow engine