SMB AI / Zeelume
Every inquiry answered before the phone rings twice
AI intake for a small business: inquiries answered, qualified, and summarized before a human picks up.
- Next.js
- TypeScript
- LLM APIs
- Workflow Automation
SERVICE / 02
AI & Automation
Working AI inside your business — intake, support, and back-office, with a human handoff.
- LLM APPS
- WORKFLOWS
- INTEGRATIONS
DELIVERABLES
What you get
Concrete things that exist when we're done — not aspirations.
- An assistant grounded in your documents, policies, and pricing
- Automation for the repetitive work: intake, triage, scheduling, follow-ups
- Integrations with the tools you already use — email, CRM, calendars, Stripe
- Written guardrails: what the AI may say, what it must never do
- A visible human-handoff path for every conversation
- Usage reporting so you can see what it handled and what it escalated
PRICING
How it's priced
Fixed scope, sprints, or retainers — always inside published bands: Under $5k, $5–15k, $15–50k, $50k+.
Pilot
$5–15k
One workflow, in production, with real users — enough to judge with evidence instead of a demo video.
Production rollout
$15–50k
The pilot hardened and extended across your business: more workflows, more integrations, monitoring, and guardrail reviews.
Care retainer
Under $5k / month
We watch it, tune it, and adjust prompts and workflows as your business changes.
PROOF
Work in this family
Products in production that came through this door.
SMB AI / HavenTech
Our own front door runs on the product we sell
Haven's own platform: AI intake, client portal, and delivery operations — the first customer of everything we build.
- Next.js
- TypeScript
- PostgreSQL
- LLM APIs
- Stripe
FAQ
Fair questions
Will the AI talk to my customers unsupervised?
Only within written guardrails you approve, and every conversation has a human-handoff path. Anything outside the guardrails escalates to a person.
What happens to our data?
Your data stays in your accounts. We document exactly which providers see what, and we configure vendors so your data is not used for model training where the provider supports it.
What if the AI gets something wrong?
We design for it: constrained answers, citations to your own documents where possible, escalation on low confidence, and logs you can review. The goal is a useful assistant, not an oracle.
Do we need to be technical to run it?
No. You get a plain-language runbook, and the care retainer exists precisely so you never have to touch a prompt if you don't want to.
Tell the Brain what you need. Get a plan in 48 hours.
Ten minutes of conversation, a structured brief, and a fixed-scope proposal with a real number on it.
Start a project