Skip to content

Voice / chat / workflow agents

Production AI agents, added to the product you already have.

We build voice, chat and workflow agents into your product and operations: governed, monitored and live in 2–3 weeks, at a fixed price.

Book a free health check

System 01 / Living architecture

Find qualified enterprise leads

Illustrative input

Agent / Data

WEBCRMDATABASEAPIEMAILDOCS

Draft outreach

Ready

Routed to a named approver

24 drafts, nothing sent

Illustrative output

Illustrative system trace / 8F4A

INPUT42ms
REASONING184ms
WEB312ms
CRM96ms
ACTION141ms
⏸ HELD FOR APPROVALDEMO

System log / illustrative

14:32:01 request.received

14:32:02 agent.plan.created

14:32:02 tool.web.selected

14:32:03 crm.record.created

14:32:03 policy.approval.required

14:32:03 outreach.held

Select a node to inspect how the architecture works.

Permissions · Policies · Human approval · Guardrails

How we work with you

Start with a free review. Ship one agent at a fixed price.

01

Free Agent Health Check

A 30-minute review of your current agent, or of the workflow you want automated, with written findings.02

AI Agent Sprint

One agent or AI feature in production in 2–3 weeks, with evals, monitoring and handover.03

Care Plan

We run, monitor and improve what we built.04

Dedicated AI Pod

An embedded team that keeps shipping.
What we build

Agents that answer, act and update your systems.

01

Voice AI agents

Inbound and outbound phone agents on Twilio, Vapi, Retell or a custom stack, tuned for low latency.

02

AI integration

Assistants, summaries and actions inside your SaaS, wired to your APIs.

03

WhatsApp and web chatbots

Agents that answer, book and update your CRM.

04

RAG knowledge assistants

Answers grounded in your documents, with sources.

05

Workflow automation

n8n and custom pipelines that remove manual steps.

06

Browser and computer-use automation

Data entry into portals that have no API.

Concept / reference architecture

Systems we build.

Practical architectures for the work businesses are trying to automate now.

Concept / reference architecture

Industry

B2B SaaS / Professional services

The gap

Sales teams lose hours to prospect discovery, account research, contact enrichment, personalized outreach and CRM administration.

Reference architecture

Lead sources
Web scraper
Data enrichment
Research agent
Lead scoring
Personalization
CRM
Outreach
Human review

What Cognizec builds

Prospect discovery
Company research
Contact enrichment
Lead scoring
Personalized messaging
CRM synchronization
Approval gates

Intended outcome

Reduce repetitive prospecting work while giving sales teams better researched, continuously updated opportunities.

How this system works

Control layer

Autonomy with a paper trail.

Every autonomous action walks the same path: an identity that owns it, a policy that permits it, a verification that confirms it, and a log that records it. Where policy refuses, or the stakes justify it, a named person decides before anything runs.

Selected stage / 03

Policy

Rules decide whether this identity may take this action, on this data, right now. Permission is checked before the call, not after it.

Enters fromIdentity

Continues toAction· permittedHuman approval· refused, or the stakes are too high

audit.logWithin policy
  1. +0.00srequestreceived · agent:intake
  2. +0.47sidentityverified · agent:intake
  3. +0.94spolicyallowed · within limit
  4. +1.41sactionexecuted · scoped call
  5. +1.88sverifypassed · as expected
  6. +2.35sauditwritten · request closed

Why Cognizec

01

Fixed prices, agreed before we start.

02

A fully remote engineering team: senior AI engineers at a fraction of UK contractor rates.

03

Guardrails from day one: permissions, human approval on risky actions, audit logs and monitoring.

04

You own the code and every account.

Technologies we work with

Built on the right stack.

We select models, tools, data systems and infrastructure around the problem—not the other way around.

OpenAI → n8n → Supabase → Next.js → Cloudflare

01Models
02Orchestration
03Data
04Applications
05Infrastructure

No single model. No fixed stack. The architecture follows the problem.

Reference architecture

Questions

How much does an AI agent cost?

A fixed price for one agent or AI feature, quoted for your workflow, plus model and telephony usage.

How long does it take?

Two to three weeks for a sprint.

Do you work with agencies?

Yes. We build white-label under your brand.

Which platforms do you use?

OpenAI, Anthropic and Gemini; Twilio, Vapi, Retell and ElevenLabs; n8n and LangGraph; AWS and Google Cloud.

Who owns the code?

You do, from the first commit.

Start a project

What should your business stop doing manually?

Tell us what takes too much time, requires too many people or doesn’t scale. We’ll design the system around it.

01

Say what the work is.

The task, roughly how often it runs, and who does it today. Three sentences is enough — the brief is the part we read first.

02

It arrives as an email.

Straight to our queries inbox, with your address as the reply-to. There is no sequence behind this form and nothing automated reads it.

03

No client names, on purpose.

The work we have done is under NDA, which is why this page argues with architecture instead of logos. Everything drawn above is a reference architecture, not an account of a delivered project.

Security check