User guide
AI Studio
In AI Studio you define your agents, manage knowledge bases, test changes before they go live and monitor performance. Viewing needs ai:read; changes need ai:write.
#Agents
Create an agent with New agent. The agent page has seven tabs:
| Tab | Contents |
|---|---|
| Configuration | Persona and instructions, tone, supported and default languages, AI disclosure text; hand-off rules: maximum AI turns, hand off when the knowledge base has no answer, escalation keywords and the hand-off message. |
| Inference | Fast, reasoning, embedding, reranker and safety models; complexity threshold, temperature, maximum reply tokens, context messages, summarise-after, retrieval top K and minimum score. |
| Tools | Which tools the agent may call (e.g. reading account data from the operator API). |
| Knowledge | Knowledge bases attached to the agent. |
| Versions | Save a draft, publish with a change note, history and configuration comparison. |
| Deployment | Websites the agent runs on, rename, enable/disable. |
| Simulation Lab | Test chats before publishing, with a trace view of every reply. |
#Knowledge
Create a knowledge base and add documents by uploading a file (txt, md, html or pdf) or by pasting text. A document added with an existing title becomes a new version of it. Agents use changes once the knowledge base is published. You can archive documents, view a document’s chunks and run a test search to see what content is found.
#Evaluations
Evaluation suites run an agent version against a set of test cases. Cases are edited as JSON; you can import the platform’s regression suite. Results show the pass rate, p95 latency and per-case outcomes. Run them before publishing a new version.
#Cost & latency
Shows runs, token usage and tokens per second per model. Monthly AI token totals are also on the Usage & billing page.