TradesKit AI

Four AI loops, all real.

Every TradesKit AI feature is a closed loop with a named trigger, a named agent run, and a named action on a specific entity. No chat overlay. No CRUD bolt-ons. Real loops you can audit.

Start free trial Meet the agents

TRIGGEREvent firesA named event fires on a specific entity

AI AGENT RUNAgent reads contextThe agent reads bounded context

AI AGENT COMMANDNamed commandThe agent issues one named command

ENTITY STATE CHANGEEntity state setEntity state changes; AIAgentRun row written

Output of stage 04 becomes the next trigger — the loop closes itself.

TRIGGER A named event fires on a specific entity

How AI usually ships

Most “AI” is a chat box next to your database.

A copilot can suggest. It can’t change state, and it leaves no record. A loop does both — every time, on a named entity, with an audit row.

Chat-on-CRUD

A copilot beside the record

You open a copilot panel
You paste in the context by hand
It returns a suggestion in prose
You copy the answer back
You update the record yourself

> open copilot> paste visit context …< "You could assign Maria Chen."> copy answer> edit Visit by hand

No state change. No audit trail.

Loops

A closed loop that changes state

An entity event fires — Visit unassigned
The trigger matches a bound agent
An AIAgentRun opens for the run
The agent reads only its bounded context
It issues one named command
The entity’s state changes
The AIAgentRun row closes — input, model, reasoning, command

trigger Visit.unassigned @ 06:30run AIAgentRun#9f3 opencommand SuggestVisitAssignment( V-4172, tech=Maria Chen)state Visit.assignment = setaudit AIAgentRun#9f3 closed

Entity state changed + audited.

The four agents

Each agent is one loop you can name end to end.

DispatchAgent

Suggests assignments before the dispatcher sits down.

At 06:30 the board is still cold. DispatchAgent has already proposed who takes what — with the proximity and skill match written into the rationale.

TriggerVisit unassigned at 06:30 (schedule event)
RunReads locations, skills, routes
CommandSuggestVisitAssignment(visit_id, tech_id, rationale)
StateVisit.assignment set

TRIGGERVisit unassigned at 06:30

AI AGENT RUNReads locations, skills, routes

AI AGENT COMMANDSuggestVisitAssignment()

ENTITY STATE CHANGEVisit.assignment set

TRIGGER Visit unassigned at 06:30 (schedule event)

Visit V-4172AI suggestedCrestview HVAC · 62 techs

Suggested techMaria Chen

Rationale1.8 mi · HVAC-Diagnostics level 4

Decided in3 seconds

VoiceReceptionistAgent

Answers after-hours calls and books the qualified ones.

The line that used to ring out at 9pm now captures the job. Structured voice turns become a BookingRequest your office confirms in the morning.

TriggerInbound call outside business hours or office line busy
RunStructured voice turns capture the caller
CommandCreateBookingRequest(...)
StateBookingRequest created; SMS confirmation sent

TRIGGERAfter-hours call

AI AGENT RUNStructured voice turns

AI AGENT COMMANDCreateBookingRequest()

ENTITY STATE CHANGEBookingRequest created

TRIGGER Inbound call outside business hours or office line busy

BookingRequest BR-2087BookedWatershed Plumbing · 41 techs

Call captured21:08

Booked windowNext morning

Confirmed by office07:02

WorkSummaryAgent

Drafts the invoice narrative when the tech taps complete.

The tech closes the visit; the customer-facing write-up is already on the invoice. Notes, photos, parts, and form responses become two plain paragraphs.

TriggerVisit → complete
RunReads notes, photos, parts, form responses
CommandGenerateWorkSummary(visit_id)
StateSummary on invoice + customer email

TRIGGERVisit → complete

AI AGENT RUNReads notes, photos, parts

AI AGENT COMMANDGenerateWorkSummary()

ENTITY STATE CHANGESummary on invoice

TRIGGER Visit → complete

Visit V-3994CompleteBrennan HVAC

EquipmentTrane XR16 — condenser fan motor

Labor2 hrs · $480

Parts$312

Drafted summaryReplaced the condenser fan motor… System tested and recharged to spec. Expected life: 8–10 years.

LeadScoringAgent

Scores every new booking 0–100 before it hits the queue.

Intake doesn’t arrive flat. Each BookingRequest carries a priority score and routing the moment it’s created — from the web form or the voice agent.

TriggerBookingRequest created (web or voice)
RunReads booking, customer history, vertical, capacity
CommandScoreBookingRequest(...)
StateBookingRequest.ai_priority_score set

TRIGGERBookingRequest created

AI AGENT RUNReads history, capacity

AI AGENT COMMANDScoreBookingRequest()

ENTITY STATE CHANGEai_priority_score set

TRIGGER BookingRequest created (web or voice)

BookingRequest BR-3310Scored 87Hale Electrical · 38 techs

SourceWeb form · 11:42

ai_priority_score87 / 100

SignalsExisting customer · replacement panel · same-day window open

How we built this

Bounded, audited, escalated.

The loop is only trustworthy if you can see its edges. Three properties hold for every agent run.

Bounded

Every agent has a defined entity scope. DispatchAgent touches assignments; WorkSummaryAgent touches summaries. No agent reaches past its boundary.

scope: one entity family per agent

Audited

Every run writes an AIAgentRun row carrying its input, the model, the reasoning, and the command it issued. The run is the receipt.

AIAgentRun { input, model, reasoning, command }

Escalated

Per-agent toggles decide what happens after the command: auto-act, or queue for a human to approve. You set the line for each agent.

mode: auto-act | queue for approval

Agent runsAIAgentRun · today

Time	Agent	Command	Entity	Result
06:30:02	DispatchAgent	SuggestVisitAssignment	V-4172	Applied · auto-act
11:42:09	LeadScoringAgent	ScoreBookingRequest	BR-3310	Applied · auto-act
14:55:13	WorkSummaryAgent	GenerateWorkSummary	V-3994	Queued for approval
21:08:44	VoiceReceptionistAgent	CreateBookingRequest	BR-2087	Applied · auto-act

TradesKit · AI run review

This is the actual surface in the product. Every run is a logged, reviewable row — the ones that need a human land in this queue; the rest are auto-applied and audited.

Run the four loops on your own shop's data.

Start a free trial and watch the first AIAgentRun row land. See the published per-tech price before you talk to anyone.

Start free trial See pricing

Two-week implementation included. Bring your own Stripe, QuickBooks, Twilio, and SendGrid. Full data export anytime.

Four AI loops, all real.

Every TradesKit AI feature is a closed loop with a named trigger, a named agent run, and a named action on a specific entity. No chat overlay. No CRUD bolt-ons. Real loops you can audit.

TRIGGEREvent firesA named event fires on a specific entity

AI AGENT RUNAgent reads contextThe agent reads bounded context

AI AGENT COMMANDNamed commandThe agent issues one named command

ENTITY STATE CHANGEEntity state setEntity state changes; AIAgentRun row written

Output of stage 04 becomes the next trigger — the loop closes itself.

TRIGGER A named event fires on a specific entity

Time

Agent

Command

Entity

Result

06:30:02

DispatchAgent

SuggestVisitAssignment

V-4172

Applied · auto-act

11:42:09

LeadScoringAgent

ScoreBookingRequest

BR-3310

Applied · auto-act

14:55:13

WorkSummaryAgent

GenerateWorkSummary

V-3994

Queued for approval

21:08:44

VoiceReceptionistAgent

CreateBookingRequest

BR-2087

Applied · auto-act