Run your first decision workflow

Start with a synthetic support ticket and inspect how typed answers become an application decision. The default example makes no network requests and needs no API key.

For help choosing or adapting an example to your own workflow, use the Awesome Jev Guide skill.

1. Run locally

Install Python 3.10 or newer, clone the repository, and run:

git clone https://github.com/AppitStudio/awesome-jev.git
cd awesome-jev
python3 examples/run.py support-routing --mock

If you already have a checkout, run the final command from its root. There are no Python packages to install.

The output identifies mock mode and shows a decision computed from a synthetic response. Open the support-routing guide to inspect its questions, fixture, policy, and expected behavior. The model values in that fixture are authored examples, not recorded Jev responses or evidence of accuracy.

2. Inspect the request

python3 examples/run.py support-routing --show-request

The request contains state, a model ID, and named questions. This workflow combines a Choice for the support queue and a Noul for explicit urgency. Each question evaluates the same state independently. Code reads the named answers and applies the review policy. The quality-rubric example introduces Score.

Try the other workflows:

python3 examples/run.py quality-rubric --mock
python3 examples/run.py span-selection --mock
python3 examples/run.py rag-triage --mock

Read Designing a Jev decision when adapting the questions to your own task.

3. Opt into a live request

Obtain access and a key through the TypeSafe console. Configure TYPESAFE_API_KEY in your shell through your usual secret-management method, then run:

python3 examples/run.py support-routing --live

Live mode sends the example's state and questions to TypeSafe and can incur usage charges. It needs network access and an authorized key. It prints the resulting decision without performing the downstream action. A live answer can differ from the synthetic fixture.

The examples default to the documented version jev-1.13.0. Consult the example configuration for overrides and the current models page before changing it. Recheck your policy when changing model versions or question wording.

4. Build on the right layer

The small HTTP client in these examples exists to make the request and failure paths inspectable. For an application, start with the official Python SDK or JavaScript SDK, which document typed interfaces and retry configuration.

Replace the sample state with a deliberately selected input, give each question one concrete job, and define review/fallback behavior before connecting a result to an action. Use labeled data to evaluate the whole policy. Running a mock fixture successfully verifies the program path; it does not validate a threshold for your workload.

For repository development, follow the checks in CONTRIBUTING.md. Node.js is needed for those development checks, not for running the Python examples.