Skip to main content
Charlie is the AI assistant inside telli. You tell Charlie what you’d like to change, it drafts the edit, and you decide what to keep. You’ll find Charlie in the chat panel on the left of the Builder. Anything you might ask a human prompt writer to do, you can ask Charlie.

Quick Start

1

Open your agent and start interacting with Charlie

Go to Agents in the sidebar, pick the agent you want to edit, and click into the Charlie chat panel on the left side of the Builder.
2

Tell Charlie what you'd like to change

Type a short message describing what you want. You can be specific or general, both work.
3

Review the edits Charlie proposes

Charlie’s changes show up inline in the prompt. Additions and removals are highlighted so you can see exactly what’s about to change.
4

Accept or reject each edit

Approve the ones you want to keep. Reject the rest. You can also accept or reject everything in one click.

What Charlie can help with

Charlie can help across multiple areas of your agent.
Tell Charlie what to change in the agent’s prompt and it drafts the edit for you.
  • Rewrite or restructure a specific section
  • Apply a tone change across the whole prompt, like making the agent friendlier or more confident
  • Insert the right variable or tool reference without having to look up the syntax
  • Adapt the prompt based on a transcript or example you share

How to talk to Charlie

Charlie understands both specific instructions and general feedback. Use whichever fits the change you have in mind.
Useful when you know exactly what you want changed.
“Shorten the greeting and add a goodbye line at the end.”
“Replace the word ‘reservation’ with ‘booking’ everywhere in the prompt.”
“Add a step asking the caller for their order number before transferring.”
If you’re not happy with what Charlie did, reject the edits and try again with a clearer instruction. You can also roll back to a checkpoint to get back to a previous version.

Files and voice input

The chat takes more than just typed text.
  • File uploads. Drop a transcript, a script document, or a sample message into the chat. Charlie reads it and uses it as context for the next edit.
  • Voice input. Tap the microphone to record a longer instruction instead of typing. Useful when you’re walking through a thought and don’t want to break flow.

Reviewing Charlie’s edits

When Charlie makes changes, they appear inline in the prompt. You decide what makes it in.
  • Additions and removals are clearly marked so you can scan the diff
  • Accept or reject each change on its own, or apply them all at once

Checkpoints

Charlie creates checkpoints automatically as it works. If you don’t like where things are going, you can roll back to any earlier checkpoint in one click.

When to use Charlie vs edit directly

Charlie isn’t a replacement for editing the prompt by hand. Some changes are faster to do yourself.
Use Charlie when…Edit directly when…
You’re restructuring or rewording a sectionYou know the exact words you want
You want a general improvement applied across the promptYou’re making a small surgical tweak
You’re working from a transcript or example documentYou’re changing a single line or word
You want help finding the right variable or tool referenceYou’re confident with the syntax

Best practices

  • Bring real call transcripts into the chat so Charlie can refine the agent based on conversations that actually happened, not just imagined cases.
  • Use checkpoints when experimenting. It’s faster to roll back than to manually undo a stack of changes.
  • Accept changes a few at a time if you’re not sure about all of them. Reject the rest and ask Charlie to try again with more guidance.
  • Start with Charlie on new prompts. Have it draft a structure for you, then refine the details by hand.

Good to know

  • Charlie never publishes your agent. Changes stay as a draft until you click Publish.
  • Charlie can also access and edit other agents if you ask it to. This is great for applying changes to multiple agents at once
  • The chat remembers what you said earlier in the same session, so you can keep iterating without re-explaining context.