Gala answers your callers from the knowledge base you give it. This guide covers what to put in it, how to write it, how to keep it current and how to test it.
An AI phone agent is only as good as what you give it. Gala does not guess: it answers from your knowledge base and hands off the rest.
Most teams already have the material. It sits in a user manual, a help center and a few people's heads.
The work is turning it into answers a caller can follow by ear. Below is the method, step by step.
Start with the questions your callers actually ask. Each item below is one part of the knowledge base.
You don't train it in the technical sense: you give it a knowledge base. Your services, your rates, your service areas, your emergency policy, the way you name things. You enter it in the platform, guided step by step, and that version governs every answer Gala gives.
Yes, including what it's not allowed to say. You set the greeting, the tone, the qualifying questions and the limits: no firm pricing, no delivery-date commitments, no technical advice outside scope. Those guardrails matter as much as the answers themselves.
Do not write the knowledge base as pairs like "if a caller asks this, answer that." Callers never use your exact words, and every new wording needs a new pair.
Write it the way you would brief a new employee: the product, how each task works, and the policies. Gala reads the whole document and finds the answer, however the caller phrases it.
A plain Markdown file works well. It is ordinary text with a few marks: # for a heading, numbers for steps, a dash for a list.
# ExampleApp: support knowledge base
ExampleApp is scheduling software for clinics.
Plans: Starter, Team and Business. Support hours: Monday to Friday, 8 am to 6 pm.
## Users and access
### Add a user
Only admins can add users.
1. Go to Settings, then Team.
2. Choose Invite.
3. Enter the person's email and pick a role.
The person receives an invitation email within a few minutes.
### Roles
- Admin: manages users, billing and settings.
- Staff: manages their own schedule.
## Billing
Billing is monthly or yearly. A plan change applies on the next invoice.
Explain the policy only. Never discuss a specific customer's account.
## Hand-offs
- Locked-out account: transfer to support during hours, take a message after.
- Billing dispute: message to the billing contact.
- Suspected security issue: transfer to the security contact right away.The example never lists questions. A caller who says "how do I give my receptionist access" still gets the Add a user steps.
A caller cannot scroll back. Every answer has to work when it is heard once.
Weak: "User management is available to administrators in the settings area."
Strong: "To add a user, go to Settings, then Team, and choose Invite. Enter their email and pick a role."
Gala hands off what needs a person. It transfers to your team or a named employee, or takes a message. The named person gets a text and an email right away.
For that to work, the rules must be explicit. Name the issue, the person, and what happens when nobody answers.
The most common way an AI agent goes wrong is a stale knowledge base. A price changes, a menu moves or a feature is renamed, and nobody updates the source.
Gala keeps answering from the old version. Callers get steps that no longer work, and your team stops trusting the answers.
Four: connecting the AI with an incomplete knowledge base, assigning nobody to validate it, automating everything at once instead of starting with inbound calls, and never listening to the first month's recordings. The fourth is the costliest — that's where you find the five questions your customers ask that nobody anticipated.
Because the knowledge base goes stale. Prices change, a service is dropped, a territory is added — and nobody tells the tool. The AI keeps answering from the old reality, errors pile up, the team stops trusting it. A 30-minute quarterly review prevents the whole scenario.
Before you forward your support line, call it yourself. Then ask a colleague who did not write the documentation to do the same.
Software companies and IT service providers get the most how-to calls. Their callers need exact steps, and their teams lose real time reading the manual aloud.
It is the material the agent answers from: your how-to steps, policies, hours and escalation rules. Gala answers from the knowledge base you give it, and that version governs every answer.
Yes. You give Gala your user manual and documentation, and it answers how-to questions from them. Answers get better when sections are short, task-based and use your exact menu names.
Whatever you list. Typical examples are security incidents, legal questions, refund exceptions and anything that needs access to a customer's account. Gala hands those to your team or takes a message.
With every product change customers can see, plus a review each quarter. A stale knowledge base is the usual reason an AI agent starts giving wrong answers.
Not natively. Gala does not create tickets or look up accounts. If your tool has an API or a webhook, our team can usually connect it.
Your documentation already holds most of the answers. Hear Gala give one. Hear Gala answer · See pricing
AI Receptionist · AI Admin Assistant · Pricing · How it works · Call Gala · Industries · Use cases · FAQ · Blog · About · Contact
Contact: contact@gavion.ai · +1 (514) 534-1298 · +1 (855) 505-0607 (toll-free) · Québec, Canada · Service across Canada in English and French.