> For the complete documentation index, see [llms.txt](/llms.txt).
> A full single-fetch corpus is available at [llms-full.txt](/llms-full.txt).
---
title: Improve with AGNT5
description: Use production run data to create datasets, score outputs, compare candidates, and gate changes.
last_verified: 2026-10-09
---

Improve covers the eval loop for deployed AGNT5 components. You use production runs to curate **[datasets](/docs/improve/datasets.md)**, score outputs with **[scorers](/docs/improve/scorers.md)**, and compare candidate prompts, models, or agent behavior in **[experiments](/docs/improve/experiments.md)** before rolling changes forward.

## Improve guides


  - [Datasets](/docs/improve/datasets) — Collect representative inputs from production runs or authored examples.
  - [Scorers](/docs/improve/scorers) — Decide whether an output meets the target behavior with built-in or custom checks.
  - [Experiments](/docs/improve/experiments) — Compare a candidate prompt, model, or agent against a baseline and gate CI on the result.
  - [Online evals](/docs/improve/online-evals) — Score sampled production runs and alert when the pass rate drops.
  - [Batch eval](/docs/improve/batch-eval) — Run evaluations from code with client.eval() and client.batch_eval().
  - [Quality cases](/docs/improve/quality-cases) — Track production failures and regressions from discovery to verified fix.
  - [Prompts](/docs/build/prompts) — Version prompt changes and compare them before they reach production traffic.


Outcome: you can decide whether a change improves agent behavior using the same platform that runs the agent.

## Next steps



Start with the data you already have from production runs.

- [Build a dataset](/docs/improve/datasets): Curate test cases from production runs and publish a version.
- [Run an experiment](/docs/improve/experiments): Score a candidate against a baseline on the same dataset.
- [Back to Run](/docs/run/overview): Capture the production runs that feed improvement work.