Skip to main content

Getting started with Lora, our AI agent builder

How to use Lora to change your AI agent's prompt and knowledge by chatting.

Written by Martha Franklin

Lora is an AI assistant built into your AI agent's settings. Instead of editing prompts and knowledge by hand, you tell Lora what you'd like to change in plain language. Lora proposes the changes and you decide whether to apply them.

Lora is only available to admins, not to supervisors or members in the workspace.

Where to find Lora

In your AI agent settings, you'll find Lora on the left side of the screen under "Copilot".
​


You can also open Lora from these places:

  • In the inbox, click "Improve AI agent" on a message from the AI agent.
    ​


    ​

  • In an AI call, click "Improve AI agent" on the summary screen
    ​

What Lora can do for you

Adjust behaviour

Describe how your AI agent should behave, for example "Be more formal" or "Always ask for the customer number before booking an appointment". Lora rewrites the relevant part of the prompt. For AI agents on the phone, Lora can also change the greeting. If Lora leaves the greeting empty, the caller speaks first.

Maintain knowledge

  • Add, edit or remove Q&A lists

  • Add a website as a knowledge source, sync it again or disable it

  • Add or remove uploaded documents as knowledge

Investigate wrong answers

Lora can read conversations from the test environment (including attachments), real inbox conversations and AI calls, and for calls that includes the transcript, forwarding and the knowledge used. Just ask "Why did the agent say this?". Lora looks at the conversation and suggests a fix.
​

Work with files and screenshots

Add files or screenshots to the chat to show what you need.
​

Approving changes

Lora never changes your AI agent without you seeing it first. Each suggestion appears as a card in the chat. There you can:

  • Save: click "Save changes" and the change is applied.

  • Reject: nothing changes.

Important: if you send a new message while a suggestion is still open, it's discarded. So save it first if you want to keep it.
​

Auto-approve: for types of changes you trust, you can turn on auto-approve. Those changes are then applied straight away.
​

No undo: there's no version history yet. That's why Lora tells you with every suggestion what it replaces. If you want an earlier version back that appears in the current conversation, Lora can restore it.
​

Tips for better results

  • Give feedback directly on the message instead of pasting the chat. When you give feedback on a message in the test environment or the inbox, Lora also sees which knowledge the agent retrieved and which tools it used. With a pasted chat, Lora only sees the text.

  • Test in the Playground. After every change you can check in the test environment whether it works. Your AI agent doesn't need to be active for this.

Going live

As long as your AI agent isn't active, it won't reply to any real customers. Activating it is a separate step that Lora can't do for you yet. When you're happy with the result, Lora sends you the link to activate it for chat or phone.
​

What Lora can't do yet

  • Configure tools (e.g. HTTP requests, call forwarding)

  • Set the voice and caller languages

  • Set up automations

  • Activate your AI agent

You'll still need to set these up yourself.

Did this answer your question?