Skip to content

Advanced agent switches (super-admin)

These switches sit in the Advanced (super admin) card of the agent page. Only a super_admin can see or change them. They are per agent, so you can try one on a single agent. The first three are on by default; the last one is off.

Reuse confirmed spellings in the email

When the caller's email contains their name or company and they did not spell that part, the agent uses the spelling they already confirmed instead of what it heard. A spelling the caller spells out for the email always wins. This never invents anything: it only reuses a spelling the caller confirmed on the same call.

Turn it off to make the agent use only what it heard for the email.

Tuned spelling-accuracy rules

Adds stricter rules so the agent does not blend a spelled name with the version it heard first. Example: it hears "Visaka", the caller spells V-I-S-O-C-K-A, and the agent must store "Visocka", not a mix. It also switches on a small extra check of spelled values by a fast AI model, so each captured spelling costs one small extra model call.

Turn it off to go back to the plain rules for this agent.

Dictation mode

When a caller spells an email, name, company, address or postcode, the system itself collects the letters across turns, reads the result back, and takes the yes or no. The AI model no longer decides when a spelling is finished or asks the caller to spell it again.

Turn it off to hand that job back to the model.

Runtime asks the next question

After the agent records a caller's answer, the runtime itself says a short "Thanks." and then the agent's next configured question, instead of making a second AI call to write it.

  • Cheaper and faster: one AI call less per answered question, roughly one to two seconds sooner.
  • Fixed wording: the acknowledgement is always the same short phrase.
  • Safe fallback: anything unusual (the caller asks something, changes an earlier answer, interrupts) still goes to the AI as before.
  • Off by default. Super-admin only. Turn it on for one agent, make a test call, and compare how it sounds before using it more widely.
  • Catalog & order taking — Let an agent answer questions about your products and take orders, and what each catalog switch does.
  • Configuring an agent — Set an agent's persona, voice, language, greeting, knowledge base, and intake questions.
  • Voices & languages — Choose how your agent sounds, preview voices, and understand language support.
  • Call forwarding & transfers — Set the conditions under which an agent hands a call to a human, and understand exactly how the transfer behaves.
  • Conversation design — Choose how an agent runs a call — simple intake, Questions & conditions (scenario), or a Decision tree.