Skip to content

AI Chatbots And Assistants That Never Guess.

Most of what your customers ask has already been answered somewhere: in a help centre, a policy PDF, a thread nobody can find. We build assistants that answer from your own material, in your customer’s language, and hand over to a person the moment they should.

At a glance Live
Timeline 4—10 Weeks
Team 1—2 People, 2—3 AI Agents
Tools & Platforms OpenAI · Claude n8n · Next.js
Deliverables Trained Assistant, Admin, Handover Docs
PythonLangChainpgvectorPostgreSQLWhatsApp BusinessZendeskIntercomSlackn8nWordPress
What we build, most requested first

Website support assistant

Answers from your help centre and policies, cites the page it used, and opens a ticket when it cannot.

Typical: most first-line questions

WhatsApp and messaging bots

The same assistant on the channel your customers already use, with order lookups and appointment changes.

Typical: replies day and night

Internal knowledge assistant

Staff ask in plain language and get the current answer from your handbooks, contracts and wikis.

Typical: onboarding gets shorter

Sales and qualification

Visitors get real answers about fit and price, and your team gets a qualified summary instead of a form fill.

Typical: fewer wasted calls

Document question answering

Long contracts, tenders and specifications made searchable in a sentence, with the clause quoted back.

Typical: minutes instead of hours

Multilingual coverage

One assistant, every market you sell in, answering in the language the question arrived in.

Typical: no separate content per language
How a build runs, with a weekly demo
  1. Week 1

    Gather the material

    We collect the content it will answer from and mark what is current, what is stale and what is missing.

  2. Weeks 2—3

    One topic, answered well

    A single subject area goes live to your team, so quality is judged on real questions rather than a demo script.

  3. Weeks 4—8

    Widen scope and handover

    More topics, escalation rules, tone, and the refusals: what it must never attempt to answer.

  4. Weeks 9—10

    Launch and tune

    Live to customers, watched daily for the first fortnight, with the answer log reviewed together.

What you get

An assistant your team can correct.

The answers come from content you control, so fixing a wrong answer means editing a page, not filing a ticket with us. Prompts, tone and escalation rules are configuration, and they live in your accounts.

01

The running assistant

Deployed on your site or channel, in your own cloud account and hosting region.

02

A content pipeline

Whatever you publish gets indexed automatically. No re-uploading when a policy changes.

03

Escalation and refusal rules

What it hands to a human, what it declines, and what it must never guess at.

04

An answer-quality log

Every question, the source used and whether it escalated, reviewable without asking us.

05

Editable prompts and tone

Voice, length and formality are settings your team changes, not code we own.

06

30 days of aftercare

Included in every package. Real questions always reveal gaps the pilot did not.

Fixed-price packages in your currency
Package 01 4 weeks

Single channel

$12k

One assistant, one channel, one body of content. The honest way to test whether it answers well enough.

  • One channel, one content set
  • Escalation to a human
  • Answer-quality log
  • 30 days aftercare
Package 03 Rolling

Assistant retainer

From $3k/mo

Someone reviewing what it got wrong every month, and shipping the fix.

  • Monthly answer review
  • Content and prompt tuning
  • New topics as you add them
  • Monitoring and on-call
Questions we get from new clients
How do you stop it inventing answers?
It answers from your content and cites what it used. Where it has no source it says so and escalates rather than guessing. We agree the refusal list before launch, and the answer log shows you every time it declined.
Will this replace our support team?
No, and we would not sell it that way. It absorbs the repeated questions so your people get the ones that need judgement. The handover rules are written by you and the assistant hands over mid-conversation, not by dropping a form.
Our documentation is out of date. Is that a problem?
It is the actual problem, and we start there. Week one is spent marking what is current, and the assistant only indexes what you approve. It is common to leave a build with better documentation than the assistant.
Which languages does it handle?
It answers in the language the question arrived in, from content written in whatever language you have. You do not need a translated help centre first, though you will spot which pages deserve one.
Where does our content go?
Into your own cloud account, in the region you choose. It is not used to train a public model, and if you stop working with us the index and the content stay where they are.
Currently booking new assistant work

What do your customers ask over and over?

Send us the top ten questions and where the answers currently live. You get a fixed quote and a timeline within three working days, in your currency.