Agent Setup

Every Bolti agent is configured from a single page in the dashboard:

Dashboard → Assistants → (your agent) → Settings

The Settings page is split into six tabs, each focused on one part of the voice pipeline. The agent detail page also has top-level tabs for Preview, Knowledge Base, Logs, and Schedule — those are documented separately.

The six settings tabs

Tab	Purpose	Common edits
Basic	Persona, language, goal, guardrails, system prompt	Adjusting tone, fixing how the agent answers, adding forbidden topics
LLM	Which language model powers the agent	Switching from gpt-5-mini to a faster/cheaper/smarter model
Voice	Text-to-speech provider, voice, speed, pitch	Changing how the agent sounds
Speech	Speech-to-text provider, model, language	Improving transcription accuracy for accents or noisy lines
Tools	Function tools the agent can call mid-conversation	Letting the agent transfer calls, look up CRM data, book meetings
Phone	Phone numbers assigned to this agent for inbound calls	Wiring a number to the agent so it answers when called

The save model

Settings are edited in place — there's no draft state. The flow is:

1. Click Edit Configuration in the top right of the Settings panel.
2. Change anything you want across any number of tabs.
3. Click Save Changes to persist, or Cancel to discard.

While you're editing, the Save Changes button stays disabled until at least one field has actually changed. Saving applies the new configuration to the next call (in-flight calls finish on their old config).

Phone number assignment is separate

Assigning or unassigning phone numbers on the Phone tab does not require the global Edit/Save flow — those changes apply immediately when you click Assign Number or Unassign.

Wizard-driven vs. raw agents

Agents created with the Create AI Agent wizard store their wizard answers (persona, language, goal, etc.) on the agent record. The Basic tab shows those structured fields and regenerates the system prompt for you whenever you change one.

Agents created via API or imported from another workspace may not have wizard answers stored — in that case the Basic tab falls back to a single raw System Prompt textarea. You can author the prompt yourself with no scaffolding. Both modes produce identical agents on the runtime side.

Testing your changes

The fastest iteration loop is:

1. Edit on the Settings tab → Save Changes.
2. Switch to the Preview tab → Connect → talk to the agent in your browser.
3. If it didn't behave the way you wanted, go back to step 1.

For changes that involve real phone behavior (DTMF, voicemail, SIP routing), use Make Outbound Call to dial yourself instead of the in-browser preview.

Where to start

If you're new to Bolti, work the tabs in this order:

1. Basic — get the persona, goal, and guardrails right first. This is 80% of agent quality.
2. Voice — pick a voice that matches the persona.
3. Preview — listen to the agent and iterate on Basic.
4. Tools — only after the conversation feels right, add tools to take action.
5. LLM and Speech — change defaults only if you have a reason to.
6. Phone — last, when you're ready to take real calls.