Skip to main content

Architect

Architect is an assistant that manages your account for you. You describe what you want in plain words and it does the work: creating an agent, attaching a knowledge base, changing a voice, or answering questions about your call activity. Users often shorten the name to Arc — either works.

Nothing changes until you approve it. Every action that touches your account arrives as an approval card you read before it runs.

Open Architect​

Architect opens in a panel over whatever screen you are on. The page behind it stays where it was — same scroll position, same data — so you can ask something mid-task and carry on afterwards.

What you can ask for​

Setting agents up

  • Create an agent in a workspace.
  • Rewrite an agent's instructions.
  • Change which voice an agent speaks with.
  • Delete an agent. It is paused rather than destroyed, so it keeps its settings and call history and can be restored later.

Knowledge bases

  • Attach an existing knowledge base to an agent.
  • Remove one from an agent — the file stays in your library and on every other agent using it.
  • Delete a knowledge base entirely. This removes the file itself and takes it away from every agent at once, and it cannot be undone.

Reading your account

  • Call volume and talk time for one agent, one workspace, or the whole account, including how those calls split into outbound, inbound, and web.
  • Recent call summaries, with caller details left out.
  • Questions about how Chatley works, answered from this documentation.
note

Architect only ever works inside the account you are signed in to.

Approving a change​

When Architect proposes something, you get a card describing exactly what will happen, with three ways to answer:

ButtonWhat it does
Approve onceApplies this one change.
Approve for this chatApplies it and stops asking for that same kind of action until you close the conversation.
DenyNothing is changed.

Some cards also offer Edit, which lets you correct a value before approving — saving your changes asks Architect for a fresh proposal rather than applying the old one.

Two things always ask, whatever you chose earlier:

  • Live agents. An agent with a phone number attached is taking real calls, so a change reaches callers the moment you approve it. The card says so plainly.
  • Deletions and undo. Neither is ever applied automatically.

Cards expire after 10 minutes. If one lapses, just ask again.

Undo​

Ask Architect to undo and it takes back the most recent change it made — the instructions it replaced, the voice it changed, the knowledge base it attached, the agent it created or deleted.

Only the last change can be taken back, and some cannot be reversed at all. Deleting a knowledge base removes the file, so there is nothing left to restore — upload it again from the Knowledge Base library. Architect tells you the reason instead of offering an undo that would do nothing.

What it asks you for​

Architect does not invent the things only you can decide. When you ask for a new agent it wants two answers before proposing anything:

  • What to call it
  • What it should handle

Everything else — type, tone, language, voice, greeting — gets a sensible default rather than a question. If your account has more than one workspace it also asks which one to use, and shows you the list rather than naming them from memory.

What it will not do​

Architect stays out of areas that need their own screen: deleting workspaces, billing, team administration, phone provisioning, campaigns, integrations, and uploading files. Ask for one of these and it says so and points you at the page that handles it.

Tips​

  • Name the agent you mean. If several match, Architect shows you the shortlist and asks rather than picking one.
  • Read the card, not just the message. The card is the exact change; the message around it is explanation.
  • Approve for this chat is useful for a run of similar edits, and it lapses on its own when you close the conversation.

Next steps​