AppBespoke · Glasgow, working US afternoons
AI agents for your business: built, run and logged, for a fixed price
Most "AI automation" stops at a demo. We build agents that do a real job inside your systems, with read access first, write access only for the actions you approve, and a log of everything they did and declined. Describe the job and get a written fit check and fixed quote by the next business day. No sales call unless you want one.
- Glasgow, UK
- Working your afternoon
- Written replies
- Code in your repo
Pilot $4,500 fixed · full builds from $9,000 · run from $600 a month
What they do
Jobs we build agents for.
If your job isn't here, describe it. The fit check is free. How the fit check works
Email and ticket triage
Every message tagged and assigned before anyone opens the inbox.
Reads the shared inbox or the ticket queue, tags each message, sets a priority and assigns it.
Its limit: It never replies to a customer without your approval, and anything it is unsure of goes to a named person.
Lead enrichment and routing
New leads arrive filled in, with the right owner.
Looks up each new lead, fills in the company details and routes it to the right owner in the CRM.
Its limit: It only writes to the fields you list, and every change shows in the log with the source it used.
Order, invoice and payment reconciliation
Three systems matched, the odd ones flagged.
Matches orders, invoices and payments across your store, your accounting system and the bank export.
Its limit: It flags mismatches for a person to fix. It does not move money or edit the books.
Drafts for a person to approve
A first draft is waiting, for a person to send.
Drafts replies, quotes and documents from your templates and past examples.
Its limit: Nothing goes out until a person has read it and pressed send.
Daily data pulls
This morning's figures, in one place by nine.
Pulls figures from supplier portals, spreadsheets and reports every morning into one sheet or database.
Its limit: Read-only logins where the portal allows them, and an alert when a pull fails.
Monitoring and alerts
You hear about the failed sync before a customer does.
Watches for a sync that failed, a listing that changed or a price that moved, and posts it to Slack or email.
Its limit: It tells you what changed and where to look. It does not try to fix it on its own.
Start here, free
Free agent fit check.
If a plain automation does it, we tell you which one and charge nothing.
Tell us what the job is, which systems it touches and roughly how often it happens. A few lines are enough. No access to anything is needed at this stage.
A written reply by the end of the next US business day: whether an agent fits, what it would do, what it would never be allowed to do, which systems it needs access to, and a fixed quote.
Send for a free fit checkWhat to send
- The job, in your own words, with an example if you have one
- The systems it touches
- Roughly how many a week
- Your company website
Free, written, by the end of the next US business day. One per company domain.
Pricing
What it costs.
A fixed price to build it, a monthly fee to run it, and a free check to find out whether you need either.
Agent pilot
$4,500
fixed, 15 US business days
- One agent, one job, one channel: Slack, email, a web form, or inside HubSpot or your CRM
- Scope agreed in writing before any code
- Read access first, write access only for the agreed actions
- An action log you can read
- A test set of real cases it must pass before it goes live
- 30 days of fixes after go-live
- Code in your GitHub from day one
- Credited in full against a full build started within 60 days
Full agent build
from $9,000
fixed per scoped agent, 4 to 8 weeks
- Several jobs or systems
- Approvals and hand-off rules
- A dashboard of what it did and what it declined
- An evaluation set and regression tests
- Documentation and a Loom walkthrough
- Scoped from the pilot or from a written brief
Then, every month
We run it, and keep it working.
3 months, then monthly.
Run
$600a month
Hosting, monitoring, model usage and a monthly review
Run and improve
$1,200a month
Everything in Run, plus 8 hours of changes a month and replies within one US business day
- Hosting and monitoring
- Alerts when it fails or declines
- Model usage up to $100 a month included, then at cost, itemised
- A monthly review of the log with you
- 3 months, then monthly, with 30 days' notice
Agent fit check: free.
Free, written, by the end of the next US business day. One per company domain.
If a plain automation does it, we tell you which one and charge nothing. Fixed builds are invoiced 50% when you accept the written quote and 50% on go-live. The run fee is invoiced monthly in advance. In USD, by card or bank transfer.
Send for a free fit checkFrom LinkedIn
What people say after working with us.
He helped us re-platform our entire website, porting it to a new CMS along with a complete redesign, in an incredibly short time scale: 12 weeks start to finish.
VP Engineering at Stream
Managed the work directly
“He just got his sleeves up and figured things out to create excellent, high-value software with processes and documentation to match.”
Chief Data and AI Officer at Waracle
Managed the work directly“An exceptional ability to grasp complex concepts and domains quickly and start contributing immediately.”
Senior Engineering Manager at Zoa
Managed the work directly“Without a doubt, an engineer anyone could fully rely on.”
Engineering Manager at Comma
Worked on the same team“As committed to the quality of our codebase and infrastructure as any permanent engineer I have worked with.”
Software Engineer at Weavr
Managed the work directly“Gets things done. I always felt I could rely on him to push forward with a project under his own steam.”
Development Lead at Geo.me Solutions
Worked on the same team
How it works
From the job to a logged run.
Read-only access comes first, and nothing goes live until it has passed a set of your real cases.
STEP 01
Describe the job
Use the form. The job, the systems and a rough weekly volume are enough to start.
STEP 02
Written fit check and fixed quote
By the end of the next US business day: whether an agent fits, what it would and would not do, and the price.
STEP 03
Read-only, then scoped access
Read-only access first. Then write access scoped to the agreed actions and nothing more, with tokens you can revoke.
STEP 04
Built against real cases
Built in your GitHub and tested against a set of your real cases before it touches anything live.
STEP 05
Go live, logged
Live with the action log, alerts when it fails or declines, and 30 days of fixes.
Guardrails
What it may do, and what it may not.
The agent works within the limits you set, and every one of them is written down before the build starts.
Built on Cloudflare Workers, Durable Objects and Workflows. Claude, OpenAI or Gemini models, chosen per job. Your own systems reached through their APIs.
- Cloudflare Workers
- Durable Objects
- Workflows
- Claude
- OpenAI
- Gemini
- 01
A written list of allowed actions
What the agent may do is a written list, agreed before the build. Anything not on it is declined and logged.
- 02
Approval for anything irreversible
Sending, paying and deleting wait for a named person to approve them. You choose who.
- 03
Your data stays where it is
Your data stays in your systems. The agent reads what it needs for each run, and our hosting runs in your region.
- 04
Scoped, revocable tokens
Each system gets its own token with only the scopes the agreed actions need. You can revoke any of them at any time.
- 05
Every run readable
Every run is written to a log you can read: what came in, what the agent did, what it declined and why.
Our own
Agents we already run.
Each of these runs on a schedule or a trigger and writes to other systems, on the same stack a client agent is built on.
AI-visibility tracker
Runs every week, asks several AI engines a fixed set of buyer questions, and uses a judge model to record which brands each answer names.
- step 1Load the fixed set of buyer questions
- step 2Ask each AI engine every question
- step 3Judge model reads each answer
- step 4Record which brands each answer names
- LLM APIs
- Workflows
- Judge model
ALSO RUNNING
On every enquiry
Enquiry pipeline
Takes every enquiry on this site, stores it, routes it by service, sends the alerts and queues the ad-platform conversion.
- Workers
- D1
- Routing
- Alerts
On every purchase
Plixir
Sends purchase and lead events from the server to the ad platforms as they happen, without a browser pixel.
- Server-side events
- Ad platforms
Per prospect
Humise
Renders a personal video for each prospect: an AI presenter reads one script, saying the prospect's name and company, queued and rendered in the background.
- AI presenters
- Rendering
Fit
Who this is for, and who it isn’t for.
A good fit
- US businesses of 5 to 200 people
- A repetitive job that touches two or more systems
- Someone who can say what a correct result looks like
- Teams that want the scope and the price in writing
Not this
- A chatbot for your website: that's a different page
- A tutorial on no-code builders
- Enterprise procurement and vendor onboarding
In writing
What we won’t do.
No hourly billing
Builds are a fixed price, quoted in writing first. Running is a monthly fee, with a change allowance on the higher tier.
No unlogged actions
If the agent did something, it is in the log. There is no mode that turns the log off.
No write access by default
Every agent starts read-only. Write access is added for the actions you agree to, one at a time.
No accuracy promises
Models get things wrong. We test against your real cases and show you the results, and we do not quote a success rate we cannot stand behind.
No "replace your team"
An agent takes a repetitive job off someone. We do not sell it as a way to cut staff.
FAQ
Questions people ask.
Which model do you use?
It is chosen per job: Claude, OpenAI or Gemini, whichever passes your test set at the lowest cost. If your company requires one provider, we use that one.
Where does it run, and where is our data?
The agent runs on Cloudflare Workers, Durable Objects and Workflows, in your region. Your data stays in your own systems: the agent reads what it needs for each run through their APIs, and the log keeps only what you agree it should.
What happens when it's unsure?
It declines, logs the case with its reason, and asks a named person on your team. It does not guess on anything outside the written list of allowed actions.
Can it work inside HubSpot, Slack, Gmail, QuickBooks, Shopify or our own app?
Yes, through their APIs: HubSpot, Slack, Gmail, Outlook, QuickBooks, Xero, Shopify, Google Sheets, most CRMs, and your own software if it has an API or a database we can be given read access to.
What if the job is better done with Zapier or a script?
Then we say so in the fit check, name the tool, and charge nothing. A Zapier or Make flow, a HubSpot workflow or a short script is often the right answer.
How do you bill?
Builds: 50% when you accept the written quote, 50% on go-live. The run fee is $600 or $1,200 a month, invoiced in advance, with a 3-month minimum. In USD, by card or bank transfer.
Do you work with UK or European companies?
Yes. We are based in Glasgow, UK. Contact us with your details and the reply comes back in your currency.
Free fit check
A written fit check by the next US business day.
Describe the job, the systems and how often it happens. The fit check is free, and if an agent fits you get a fixed quote with it.
Pilot $4,500 fixed · full builds from $9,000 · run from $600 a month.