Getting Started

No installation or sign-up needed to explore devx-ai — open the preview and follow this tour section by section.

1. Start on the Dashboard

The Dashboard gives you the state of the world at a glance: average pass rate across recent evaluation runs, average latency, total trace cost, and error rate, plus a table of the most recent runs.

2. Look at a Dataset

Open Datasets and click into "Customer Support QA." Each row is a golden test case: an input question, the expected output, and the retrieval context it should be grounded in. Try adding a test case of your own with the "Add test case" button — it appears immediately in the table.

3. Run — and read — an Evaluation

Open Evaluations and click into any run. Each test case shows the expected vs. actual output side by side, plus a score for every metric (Answer Relevancy, Faithfulness, Hallucination, Contextual Precision) with a pass/fail badge.

4. Diff two runs in Compare

Open Compare, pick a baseline run and a candidate run, and you'll get a per-test-case delta table — which test cases got better, which got worse, and by how much on each metric.

5. Version a prompt and open it in the Playground

Open Prompts, pick a prompt, and add a new version. Click "Open in Playground" on any version and it'll prefill the prompt text and default model, ready to run.

6. Inspect a Trace

Open Traces and click into any row to see the full prompt/completion, token counts, cost, and a span timeline (retrieve → generate → postprocess).

7. Check Annotations, Alerts, and Analytics

Annotations is a human review queue — thumbs up/down a trace with a note. Alerts shows configured threshold rules and a feed of triggered events, with a live enable/disable toggle. Analytics breaks usage and cost down by model and route.

Ready for the details on any one section? Jump to Concepts in the sidebar.