Fine-Tune Answer Quality
This is the last step of Stage 4. Your agent already works; here you make its answers sharper by choosing the right AI model, controlling how much it reads and remembers, tuning how it searches your knowledge, and cleaning up the pieces of text it answers from.
You do not need any of this to launch. The defaults work for general use. Come here when real conversations show a pattern: answers that miss information you know is in your documents, answers that are too slow, or the wrong level of detail.
Before you start
- Your agent has content in Teach Your Agent and has been tested in Preview and Publish Changes.
- You have a handful of questions where the answer is wrong, incomplete, or too slow. Write them down: you will use them to check whether each change helps.
- Ideally, those questions are saved as scenarios in Test Automation, so you can rerun them after every change.
Pages in this step
| Page | What you tune | Where in the CMS | Reach for it when |
|---|---|---|---|
| Model & Context | The conversation AI model, what users may attach, and how much history the agent carries | Conversation → Model & Context | Answers are too slow, too shallow, or the agent loses track of long conversations. |
| Retrieval | The reference toggles, how many chunks are retrieved, matching tolerance, and reranking | Knowledge → Retrieval | The agent misses information that is in your documents, or pulls in irrelevant text. |
| Chunks | The individual pieces of text the agent retrieves | Knowledge → Chunks | You want to see exactly what the agent knows, fix a badly parsed passage, or add a short fact. |
Where to start: if the problem is a wrong or missing answer, open Chunks first and check that the right text exists and is active. Only then adjust Retrieval. If the problem is speed or depth, start with Model & Context.
Change one thing at a time. Save, test your questions in the Preview panel, and only then change the next setting. Otherwise you cannot tell which change helped.
Next steps
- Start with Model & Context, or jump to the page that matches your problem above.
- Measure the effect of your changes with Test Automation.
- Back to the big picture: Overview.