Coming soonAI voice receptionist for service businesses
Callora
A website and interactive demonstration for an AI receptionist concept.
- React
- Vite
- TypeScript
- +2 more
ServicesAI Copilots & Agents
AI agents thatwork insideyour tools.
AI agents and copilots that read your own records, draft the next step and wait for a person to approve it. Everything they do runs in your accounts and is written to an audit log.
Agent draft · Fernhill Studio
Thanks for getting in touch. Would Tuesday at 10:00 suit for a call?
Read: CRM record, 2 approved documents. Nothing leaves your accounts.
Waiting for a person.
How it works
How an AI agent works inside your business.
One enquiry, followed through the loop. The agent pauses for a person before it changes anything. Try Approve and Dismiss when it stops.
Step 1 of 6
- Trigger
- Context
- Plan
- Human approval
- Act
- Log
Agent view
Inbox · new
Fernhill Studio
Hi, could you send a quote for a new shop front? We would like to start next month.
Agent message
New enquiry from Fernhill Studio. Let me check what we already know.
Guardrails
- Scoped permissionsIt only sees the mailbox and records you grant.
- Human approval for risky actionsReplies and record changes wait for a person.
- Full audit logEach step is recorded with who approved it.
- No data leaves your accountsData stays in your own Supabase or Convex project.
Audit log
- No actions yet. Nothing happens without approval.
Trigger. A new enquiry email arrives and wakes the agent.
Who it is for
Is this for you?
- Teams who waste time searching shared drives, inboxes and old tickets for an answer.
- Businesses that want AI inside their own system rather than another standalone chat tool.
- Operations leads who want agents to prepare work, with a person still signing it off.
What we build
What this covers.
Natural-language search
Ask for a record by description, serial number or customer and get the matching item, as in our inventory demo.
In-app copilots
A side panel in your dashboard that explains a screen, summarises a record or drafts the next step.
Approval-first agents
Agents that propose changes, such as a re-planned week or a follow-up email, and apply them only after approval.
Retrieval over your documents
Policies, manuals and past work indexed so answers cite the source they came from.
Model routing and guardrails
OpenRouter lets us pick the right model per task, with prompts, limits and logging that you can inspect.
How we work
Four steps, no surprises.
- 1
Pick one real job
We start with a single task people already do, such as finding a part or answering a customer query.
- 2
Prepare the data
We connect the sources, clean what is needed and decide what the assistant may and may not read.
- 3
Build, then test against real questions
We try it on genuine examples, record where it fails and tighten prompts, retrieval and permissions.
- 4
Launch with oversight
It goes live with approvals and logs switched on, and we review its answers with you before widening its role.
Selected work
Where we have done this.
Each project is labelled Live, Demo or Coming soon, so you can see what is in use today and what is on its way.
Stack
Tools we use for this.
- OpenRouter
- Next.js
- Convex
- Supabase
- Clerk
- Vercel
- HubSpot
- Zapier-style webhooks
FAQ
Questions we get asked.
Where does our data go?
We agree where your business data is stored and which AI providers may process relevant information. We explain what information is shared and what controls are available for each project.
Can an agent act on its own?
Only where you decide it should. By default, agents suggest and a person approves, and every action is logged. Low-risk steps can be automated later once you trust the results.
What if the AI gets something wrong?
It can, which is why answers point to their source and risky actions need approval. We test against real examples before launch and keep reviewing results afterwards.
Do we need a lot of data first?
Not necessarily. Many useful assistants work on the records and documents you already have. We review what exists during scoping and tell you honestly if it is not yet enough.
Related services
Often paired with this.
- AI Workflows & AutomationSee how it worksAutomations that move data between your tools and take admin off your team.
- Custom Operating SystemsSee how it worksCustom software that brings your everyday business processes together in one place.
- SaaS DevelopmentSee how it worksSubscription software where customers can sign up, manage accounts and pay online.