Alongside Microsoft Copilot.
Then past where Copilot stops.
An AI agent is a smart, reusable workflow. It carries out one repeatable job the same way every time: reads the input, does the work, and queues the result for a person to approve. Your engineers keep the judgement. The agent takes the typing. We build them on the Microsoft platform your IT team already trusts, and local-first on your own kit where the data must never leave.
Agents we build for firms like yours
Each one is a working page in your dashboard, scoped to one job, delivered with training for the people who review its output. The audit ends in a prioritised agent register for your firm: which agents, in what order, with what approval gates.
These are patterns we have built, not shelf products. The audit determines which agents we configure for your operation and how they are governed once live, ending in a prioritised agent register for your firm. And inside your own dashboard, every agent carries whatever name your team gives it.
Engineering delivery
01SCRIBE · the case-study agent
Proven patternTakes a finished technical report and drafts the case study and bid submission from it. The job nobody ever has time for.
SCRIBE · the case-study agent
Proven patternTakes a finished technical report and drafts the case study and bid submission from it. The job nobody ever has time for.
02WARDEN · the compliance-check agent
Proven patternReads a draft against the standard it claims to meet and flags every line that does not hold, with the clause next to it.
WARDEN · the compliance-check agent
Proven patternReads a draft against the standard it claims to meet and flags every line that does not hold, with the clause next to it.
03FOREMAN · the site-report agent
Proven patternTurns the week's site notes and photos into the progress report your client actually reads, in your house format.
FOREMAN · the site-report agent
Proven patternTurns the week's site notes and photos into the progress report your client actually reads, in your house format.
04GATEKEEPER · the enquiry-triage agent
Proven patternReads inbound enquiries, pulls the relevant past projects, and hands your team a brief with a suggested response, not a bare email.
GATEKEEPER · the enquiry-triage agent
Proven patternReads inbound enquiries, pulls the relevant past projects, and hands your team a brief with a suggested response, not a bare email.
05MACHINIST · the CAD-request agent
Live on this siteTakes a plain-English part description and returns a STEP file plus the generating script, through the same engine on this site.
MACHINIST · the CAD-request agent
Live on this siteTakes a plain-English part description and returns a STEP file plus the generating script, through the same engine on this site.
06SURVEYOR · the land data agent
In developmentIn development: pulls planning, land and environmental data for a location into one brief, so feasibility starts from facts, not tabs.
SURVEYOR · the land data agent
In developmentIn development: pulls planning, land and environmental data for a location into one brief, so feasibility starts from facts, not tabs.
07ESTIMATOR · the take-off agent
In developmentIn development: reads drawings and schedules of work and drafts the quantity take-off, with every figure traceable back to the sheet it came from.
ESTIMATOR · the take-off agent
In developmentIn development: reads drawings and schedules of work and drafts the quantity take-off, with every figure traceable back to the sheet it came from.
08TIMEKEEPER · the timesheet agent
In developmentIn development: turns site presence, job notes and calendars into draft timesheets, so the week's hours land ready for approval instead of Friday-night memory.
TIMEKEEPER · the timesheet agent
In developmentIn development: turns site presence, job notes and calendars into draft timesheets, so the week's hours land ready for approval instead of Friday-night memory.
Winning work & communications
The commercial layer, run on the same spine: the same review gate sits on what reaches LinkedIn as on what reaches site.
01BIDWRIGHT · the tender-pack agent
Proven patternAssembles past performance, method statements and team CVs against the tender's own headings, then queues the pack for review.
BIDWRIGHT · the tender-pack agent
Proven patternAssembles past performance, method statements and team CVs against the tender's own headings, then queues the pack for review.
02SIGNAL · the project-content agent
In service on our own contentTurns approved project information into LinkedIn and company content that keeps the firm visible. Confidentiality and claims checked, nothing publishes unreviewed.
SIGNAL · the project-content agent
In service on our own contentTurns approved project information into LinkedIn and company content that keeps the firm visible. Confidentiality and claims checked, nothing publishes unreviewed.
03DISPATCH · the client-communications agent
In service on our own communicationsDrafts company announcements, existing-client updates and past-client letters in the firm's voice, each against the audience it is for.
DISPATCH · the client-communications agent
In service on our own communicationsDrafts company announcements, existing-client updates and past-client letters in the firm's voice, each against the audience it is for.
04CONCIERGE · the client-portal agent
Proven patternGives every client a live portal view of their jobs: progress, documents and approvals in one place, so the update phone call answers itself. Ask for the walkthrough on a call.
CONCIERGE · the client-portal agent
Proven patternGives every client a live portal view of their jobs: progress, documents and approvals in one place, so the update phone call answers itself. Ask for the walkthrough on a call.
Same governance. Three places to run.
Where an agent lives depends on what it touches. Bid text is not inspection data. So we deliver in three modes, and the audit tells you which mode fits which job.
your 365 tenant
Copilot agents, in your tenant
Microsoft-native agents configured inside your own 365 tenant. Your data stays in your Microsoft estate, under your licences and your admin controls. We configure them properly, because it is what we do full time rather than a side project for an IT manager who already has a day job.
your own hardware
Local-first agents, on your infrastructure
For regulated work. The agent runs on your own hardware, results land in your own dashboard, and the data never leaves the building. Not our servers, not anyone’s hub. Yours.
Microsoft Teams
The Slingshot app, inside your Teams
A Slingshot app your admin adds to Microsoft Teams. Your team tags an agent in the chat they already use, the agent does the job, and the result comes back for sign-off. No new software to learn. In build now, first client tenants next.
What the statuses mean: available now is a mode we build and hand over today, in build is one we are still finishing. They describe what we deliver, not a count of installs. Ask on a call and we will tell you exactly what is running where, subject to what each client lets us name.
The engineering world just started building these.
The first job firms ask us to point an agent at is nearly always the same: turning technical reports into bid submissions and case studies, because there is never enough time to pull strong ones together. It is the exact shape of work an agent is good at, which is why it comes up first.
The difference with Slingshot: you do not need an internal innovation team to get there. We configure Copilot where Copilot is strong, build local-first agents where it is not, and put a review gate in front of anything that leaves the building. And your data never goes through a third-party AI hub. That rule is the product.
We build on Microsoft Copilot where Copilot is strong, and go where Copilot cannot. Your agents start local, on your own infrastructure, in your own dashboard, so your data never leaves the building. When you are ready, they surface inside Microsoft Teams where your people already work.
Where Slingshot sitsEverything the firm knows, connected.
The operator dashboards we build ship with a brain: one navigable graph that connects the firm’s jobs, records, specs, standards and standing decisions. Click a node and fly to it. Ask a question and the answer comes back with the source pinned alongside. This is a live preview on the template’s demo dataset. Drag it.
Speak to your folders
The brain links to the SharePoint you already run. Past projects, method statements, the decision register: asking your company a question means literally asking your own files, and the file that answered is shown next to the answer.
Runs in your own estate
No third-party AI hub, no copy of your records on someone else’s servers. It runs where your data already lives, which is the reason a firm that takes data ownership seriously can have this at all. The dashboard preview on our landing page is the same product family: what we build is what that preview shows.
Five agents work in the brain. Three ship in the template, two are configured to your firm.
The graph above is not a wall chart. Five agents work in it, and because it links to your own SharePoint, asking them a question is asking your own folders.
ARCHIVIST
Files every new record into the brain and keeps the graph current, so nothing the firm produces goes missing.
RECALL
The speak-to-your-company agent. Answers questions from your own records with the source pinned alongside.
CARTOGRAPHER
Maps the clusters and surfaces connections between jobs, specs and past work you did not know were there.
WATCHKEEPER
Watches the graph for stale records, expiring certs and standards updates, and raises the flag before it bites.
ADVISOR
The senior-lens panel. Questions answered against the brain, with named sources, like asking the longest-serving engineer in the building.
The brain runs local-first on your own kit: delivery mode B in the register above, under the same governance spine as everything else on this page.
Run to a standard a board can sign off.
Every agent, in every mode, runs on the same three rules. This is the part the demos usually skip, and the part that decides whether AI survives contact with your QA manager.
A named UK supplier
You deal with a named UK company and the people in it, on a contract. Not an anonymous tool with a foreign privacy policy. Someone is accountable and you know their name.
Signed off and logged, every run
Nothing an agent produces reaches a client, a regulator or a live system without a person approving it, and every run is logged: what went in, what came out, who signed it off. When someone asks how a document was produced, you can answer.
No third-party AI hubs. Ever.
Your data never routes through an external AI hub or an unknown vendor’s servers. Firms that take data ownership seriously can buy from us because of this rule, not despite it.
AI on the tender pipeline itself.
Bid and tender writing is where engineering firms bleed senior hours, and it is where the sector's first AI tools are landing. BIDWRIGHT drafts against the tender's own headings from your past performance, and the funnel shows where every live bid sits, from qualification to submission.
Governed like everything else: a bid lead signs off every section, and the pack never leaves the building on its own.
Find the agent hiding in your workload
The Board AI Readiness Score tells you where your firm stands. A call tells us which repeatable job is costing you the most. Either way you leave with something concrete.