What is Banh?
Banh turns unstructured input into bounded decisions, then evaluates the rules you write. Define a workflow in YAML, run inference locally with Laya, and return a JSON-compatible value to your application.
How it works
Section titled “How it works”- Input: provide JSON or text, such as a support ticket.
- Decisions: ask the model to choose an option, answer a yes/no question, or score a scale. Laya evaluates all decisions in one model call.
- Flow: compare decision values or metadata in ordered rules. The first matching rule returns your output.
The flow is deterministic for a given set of decision results. Model predictions can vary; the rules make the response to those predictions explicit. Banh does not run agent loops, execute arbitrary code, or perform side-effect actions.
Choose your starting point
Section titled “Choose your starting point”| I want to… | Start here |
|---|---|
| Run my first workflow | Local quickstart |
| Write decisions and routing rules | Workflow syntax |
| Embed Banh in an application | TypeScript usage |
| Understand hosted execution | Banh Cloud availability |
| Look up a command or endpoint | CLI reference or HTTP API |
Current availability
Section titled “Current availability”Banh is in early development. The local runtime uses real Laya inference. Packages are not published yet, so the quickstart uses a source checkout.
The Banh Cloud development service supports authentication, immutable workflow versions, and recorded runs. Its execution backend currently returns fixed test answers. Hosted Laya inference is not enabled yet.
What belongs in a workflow?
Section titled “What belongs in a workflow?”Good starting points include support routing, warranty triage, and classification with a human-review path. Keep questions focused and inputs short. Laya’s English checkpoint has a 512-token context window, including question headers; long input may be truncated.