Chatbots that know what they don’t know.

We build production chatbots for support, sales, and ops — grounded in your knowledge, eval-tested before launch, and honest enough to escalate when they should.

Live · production chat · 24/7
Support chat · #widget-prodRAG-grounded · evals on
Visitor14:23

What’s your cancellation policy?

Bot14:23

30 days notice, prorated refund. Last day of access is your billing-cycle end.

Source · terms.pdf §4.2
Visitor14:24

Can I get an exception on the annual plan?

Bot14:24

Worth a quick chat. I’ll route you to your account team.

Routed · @sarah · account team
12K msgs / mo · 1.2s avg0 hallucinations
24/7always-on coverage
2 wkspilot to production
100%code you own, no lock-in
The problem

Why chatbots embarrass companies.

Two ways chatbots embarrass you. Either they escalate everything to a human, which defeats the point, or they confidently lie, which is worse. The fix isn’t a bigger model. It’s grounding answers in your knowledge, evaluating before you launch, and shipping a fallback policy that knows when to hand off. We build the kind that doesn’t get screenshot-shamed on Twitter.

Chatbot incident log · sample3 anti-patterns
01 · Fail·HALLUCINATED_POLICY
CRIT

Bot fabricated a 60-day refund policy that doesn’t exist anywhere in the company docs. Customer screenshotted it.

confidence · source:0.92 · none·generic LLM, no RAG
02 · Fail·ESCALATION_OVERFLOW
WARN

89% of queries escalated to a human. The bot became a dressed-up contact form.

deflection:11%·over-cautious prompt
03 · Fail·NO_CITATION
CRIT

Confident answer delivered without a source. Engineers can’t reproduce, users can’t verify, audit fails.

citation coverage:0%·no grounding
What we ship

Four shapes, one engineering team.

01

Support chatbots with deflection

Resolve repeat questions automatically with citation-backed answers. Escalate to a human when confidence drops.

02

Sales qualifiers

Pre-qualify leads, capture intent, and route to sales with a hot summary attached.

03

Internal ops assistants

Chatbots that know your runbooks, SOPs, and internal docs. Save your team from re-explaining the same things.

04

Knowledge base companions

Conversational search across your documentation, with the receipts. Replace "where is that doc?" Slack messages.

See case study
Why Zaibex

Side-by-side with a typical agency.

We don’t outbid the cheap shops, and we don’t pretend to be McKinsey. Here’s where the real practical difference lives.

Typical agency
Zaibex
Junior dev assigned after kickoff
Senior engineer owns end-to-end
Generic chatbot from a template
Custom-trained on your knowledge
$15K, 8-week timeline
$3K–$10K, 2–4 weeks to production
Hourly billing, scope creep
Fixed-price per phase
No evals, hope it works
Adversarial eval suite before launch
Built on a closed no-code platform
Code you own, hosted on your infra
The stack

The exact tools and why we chose them.

No mystery stack, no platform lock-in. You see what we use, you read why, and you own the keys on day one.

01 · LLM
Claude

Best at refusing what it doesn’t know — the right default for an honest chatbot.

02 · Retrieval
Postgres + pgvector

One database for vectors, BM25, and your domain. No extra vector store to maintain.

03 · Frontend widget
Custom React + Next.js

Branded to match your site. Streaming, accessible, mobile-first.

04 · Knowledge
Custom ingestion + LlamaIndex

Auto-indexed from your docs, PDFs, Notion, Confluence, or APIs.

05 · Eval framework
Custom regression tests

Adversarial test set on real questions before every deploy.

06 · Monitoring
Custom + Vercel Logs

Track answer quality, escalation rate, and what users actually ask.

The process

Discover · Architect · Build · Ship.

Four stages, named timelines, named deliverables. No open-ended discovery. No moving goalposts.

01·3 days

Discover

We audit your knowledge sources, your top 50 questions, and what good escalation looks like.

  • Top-50 questions audit
  • Knowledge source map
  • Escalation policy
02·1 week

Architect

We design retrieval, prompts, evals, and the escalation policy — and write the evals before writing the bot.

  • Prompt + retrieval spec
  • Escalation policy
  • Eval suite
03·1–2 weeks

Build

Sandbox first, then production. Iterate against the eval suite, not against gut feel.

  • Chatbot on staging
  • Eval pass rate >95%
  • Branded UI integrated
04·1 day

Ship

Cutover during a quiet hour. We’re on-call for the first 72 hours.

  • Production cutover
  • Analytics + escalation tracking
  • 72h hyper-care
In production

Real numbers, named client.

40%faster knowledge retrieval, with citation-backed answers
AI knowledge platform
Read the full case study
The problem

A legal-tech team needed users to find contract precedents fast without trusting fabricated answers. Generic chatbots hallucinated; precedent search took hours.

Our approach

We built a RAG-powered chatbot on Postgres + pgvector with strict citation enforcement, eval-driven prompt tuning, and a contract-aware UI.

The outcome

40% faster contract creation. Citation-grounded answers users can verify. Built in 6 weeks, in production today.

Pricing & timeline

Fixed price. Fixed scope. Public ranges.

We don’t hide pricing behind a sales call. Pick the tier that matches your stage. The discovery call confirms scope, not budget.

Pilot
$3K
2 weeks
  • Custom chatbot grounded in your knowledge
  • Branded chat widget on your site
  • Eval-tested before launch
  • Production deployment
  • 3 months of post-launch support
Custom
Let's talk
Tailored to scope
  • Fully customized to your operations
  • Multi-source knowledge, complex flows
  • CRM and helpdesk integrations
  • Ongoing optimization and on-call
Honest answers

The questions we actually hear on calls.

How long does a chatbot take to ship?

A focused pilot ships in 2 weeks. A multi-source production chatbot with CRM integration ships in 3–4 weeks.

Can it actually deflect support tickets?

Yes, if the questions are knowledge-answerable. Deflection depends on your knowledge base completeness and how clearly users ask. We measure deflection from week 1 and tune the answer set monthly.

How do you stop it from hallucinating?

Retrieval-first prompting with citation enforcement, an eval suite tested before every deploy, and a fallback policy when confidence drops below threshold. The bot says "I don’t know" rather than lying.

Does it integrate with our helpdesk or CRM?

Yes. We integrate with Zendesk, Intercom, Front, HubSpot, Salesforce, and any system with an API. The chatbot can create tickets, search your CRM, and pass context to humans.

Can we use it in multiple languages?

Yes. Modern LLMs handle 20+ languages well. We’ve shipped chatbots in English, Urdu, Arabic, and Spanish with language-detection routing.

Do we own the chatbot?

Yes. Code, prompts, and evals live in your account. We use Claude — you hold the API key. No platform lock-in.

Ready when you are

Good chatbots don’t pretend to know everything. They route to humans when they should and answer with citations when they can. We build the kind your users actually want to use — and your team can trust.