freestart.ai
freestart.ai · operator-built ai

We run our own business on AI. In production.

freestart.ai is the engineering arm of Freestart Plc. The AI we run our own business on — scoring calls, building pipeline, coaching the floor — is the AI we make available to yours. Built in production, not in a deck.

See the products → How we build
The AI arm of Freestart Plc — building websites and software for British businesses since 1999.
production — freestart.ai LIVE
freestart@prod:~$ status --all
callhubscoring live calls
prospecthubpipeline engine · v4
traininghubcall coaching
infra27 workloads · 2 regions
4 services up · checked just now
// the proof

Proof, not slideware.

Every number below is something we run in production today — the part a consultancy selling decks structurally can't show you.

Live
CallHub scores real sales calls in production
callhub · in production
v4
ProspectHub pipeline engine, fourth generation shipped
prospecthub · shipped
12727
VMs consolidated to 27 workloads per region — orchestrated by agentic AI
infra · wireguard mesh · zfs
123k
Local-plan policy chunks across the UK
planning · 347 of 361 · 96%
// flagship · sales-enablement stack

Score the calls. Train the gaps. Feed the pipeline.

One system, three parts — the stack we run on our own sales floor before we sell it to yours. Each piece feeds the next.

01 · CALLHUB LIVE

Score the calls

Every sales call transcribed and scored against your playbook, automatically — so coaching is based on what actually happened.

  • Live transcription & sentiment
  • Playbook-based call scoring
  • Objection & outcome tagging
View CallHub →
02 · TRAININGHUB LIVE

Train the gaps

Turns the patterns CallHub finds into targeted coaching — drills the exact moments each rep keeps losing, not generic modules.

  • Per-rep skill gap analysis
  • AI roleplay against real objections
  • Progress tracked back to scores
View TrainingHub →
03 · PROSPECTHUB v4

Feed the pipeline

Researches, qualifies and queues the next set of prospects — so a better-trained floor never runs out of the right calls to make.

  • Automated prospect research
  • Fit & intent qualification
  • Prioritised call queue
View ProspectHub →
// the loop
self-improving
every call trains the next
01 PROSPECTHUB
Feed the pipeline
02 CALLHUB
Score the calls
03 TRAININGHUB
Train the gaps

One system that compounds.

Each stage feeds the next — and the data from every call trains the models behind the one after it. The more we run it, the sharper it gets.

  • PROSPECTHUB AI researches and qualifies the next accounts, so the floor always has the right calls queued.
  • CALLHUB AI transcribes and scores every live call against the playbook — no manual QA.
  • TRAININGHUB AI turns those scores into targeted coaching, and sharper reps surface better prospect signals.
// vertical · uk planning & property

Planning intelligence, built on the actual policy.

A different problem, a different buyer. The same operator-built discipline — pointed at the UK planning system. PlanningConsult and PlanningAssist run retrieval over the real local-plan corpus, not a generic model guessing at policy.

PLANNINGCONSULT IN DEVELOPMENT

Check the constraints before you commit.

Retrieval-augmented checking across the local-plan policy corpus, with a coverage gate that flags when the evidence isn't strong enough to answer — rather than guessing.

For planning professionals

The statement tier resolves the site's LPA, pulls the governing local-plan and NPPF policy, and drafts a planning statement in the standard structure. Every point is cited to the policy it rests on, each citation verified against source, and the draft is adversarially reviewed before it reaches you — a submission-ready first draft, not a blank page.

  • RAG over the live policy corpus
  • CoverageGate — knows what it doesn't know
  • Planning-statement tier — LPA-resolved, every claim cited and citation-verified, adversarially reviewed
View PlanningConsult →
PLANNINGASSIST IN BETA

The planning question, answered from the rules.

A source-grounded chat assistant for the questions homeowners, developers and architects ask first — grounded in real policy and preverified rule cards.

For the question before the application

The assistant matches each question to a preverified rule card, then scopes retrieval to the policy that actually governs it. The answer is anchored to a source already checked against the legislation — the real limits, not an invented number — and where the corpus can't answer, the SufficiencyGate says so instead of guessing.

  • Source-grounded chat over the policy corpus
  • SufficiencyGate — answers only when the evidence clears the bar, else says so
  • 56 preverified England rule cards
View PlanningAssist →
~123k
local-plan policy chunks across 347 of the UK's 361 planning authorities (96%), spanning 818 local-plan documents — the corpus PlanningConsult and PlanningAssist reason over.
corpus · live count · 25 Jun 2026
// selected partners

We build the same systems for a handful of other companies.

Not as an agency — as operators who built this for ourselves first. We take on a small number of partners at a time, where the problem looks like one we've already solved in production.

Enquire about a build →