Custom AI agents for small and mid-sized businesses, built on Claude and OpenAI
Most small businesses don't have the engineers to build agent infrastructure: only 14% of small EU firms employ an IT specialist. Managed agent platforms now run that infrastructure for you. We build the rest: custom AI agents on Claude Managed Agents or the OpenAI Agents API, connected to your tools, with approvals, budgets and tests.
Why are managed AI agents the better fit for small and mid-sized businesses?
Because you want the agent, not the servers behind it. A managed runtime keeps the infrastructure running, so you don't need to hire engineers to keep your agent alive.
14%
of small EU businesses employ an IT specialist.
Eurostat, ICT specialists in enterprises, 2024
- 01
No engineers needed
You don't have an engineering team, and you shouldn't need one.
Only 14% of small EU businesses employ an IT specialist. Lack of expertise is the top reason businesses that wanted AI never adopted it.
- 02
Zero infrastructure
No servers to babysit.
The runtime handles the secure workspace, saved progress, logins, permissions and logs. That's the "months of infrastructure work" you'd otherwise have to build before your agent does anything useful.
- 03
Pay per use
You pay only while it works.
You pay the platform for what the agent uses, with no servers running in between. On Claude Managed Agents, active time adds $0.08 an hour; idle time costs nothing.
- 04
Runs for hours
Long and recurring jobs just run.
Your agent can work for hours, pick up where it left off, and run on a schedule.
- 05
Your own cloud
Strict data rules? We self-host.
Managed runtimes don't offer zero data retention yet. When your data can't leave your control, we run the agent on Vercel, using only AI models that keep none of your data, or on your own cloud.
Why is your team still doing repetitive work an AI agent could handle?
Because the jobs are real, but there's nobody to automate them. Most small businesses have no IT specialist, so the work stays manual and lands on the same few people every week. Anthropic and OpenAI built their managed agents for exactly these jobs. The agent platform now exists. What's missing is someone to connect it to your tools and your way of working.
Jobs a custom AI agent can take off your team
-
Triage the inbox
Sort the support inbox by topic and urgency, and send each request to the right person.
-
Draft replies
Draft replies to customer emails in your team's tone, for someone to check and send.
-
Answer stock questions
Answer order and stock questions, and send a daily digest of low stock, slow sellers and order problems.
-
Review the paperwork
Review invoices, contracts and vendor forms, and pull out what needs attention.
-
Prep every meeting
Research everyone on the guest list before a sales meeting.
-
Build the weekly report
Answer questions from your sales and finance data, and build the weekly report.
Claude Managed Agents, OpenAI Agents API or Vercel: which is right for your business?
Pick the one your team already uses. If cost control or strict data rules matter most, we run it on Vercel or your own cloud.
Claude
Anthropic runs it
Claude Managed Agents
Anthropic runs the agent for you, including long jobs and scheduled runs.
Best if you want work done in the background without anyone watching it.
OpenAI
OpenAI runs it
OpenAI Agents API
OpenAI runs the agent for you.
Best if your team already works in ChatGPT and OpenAI.
Own cloud
You run it
Your own cloud
You keep full control of where data goes.
Best if strict data rules apply to the job.
Vercel
Your Vercel account
Vercel
Runs the agent on your own Vercel account. If one AI model goes down, it switches to a backup. Spending limits apply per team, and you can use only AI models that keep none of your data.
Best if cost control and data rules matter most.
How does a custom AI agent connect to your tools, team and data?
Your agent sits in the middle. It talks to AI models, your team, the web, your business systems, other APIs and other agents, and every request in or out passes its guardrails first.
Click any box. Every request in or out passes the agent's guardrails. Tap any box. Every request in or out passes the agent's guardrails.
Connects through the agent
AI models
- Claude, GPT and others
- via AI Gateway, with fallback
Connects through the agent
Your team
- Asks in Slack, Teams or email
- Approves what matters
Connects through the agent
The internet
- Web search and page fetch
- Allowed sites only
We build this
Your custom AI agent
- Skills: your process, step by step
- Guardrails: approvals, budget per run, audit log
- Runs on Claude Managed Agents, OpenAI or Vercel eve
Connects through the agent
Business systems
- CRM, support, billing, finance
- via official + custom MCP
We build this
Other APIs
- Any tool with an API
- via custom MCP tools we build
Connects through the agent
Other agents
- Coding agents and subagents
- Hand off and collect results
Your team asks and approves. AI models do the thinking, with a backup if one goes down. Your business systems and any other tool with an API connect through official or custom connectors. The agent can search the web on sites you allow, and hand work to coding agents or helper agents and collect the results.
How do we build a custom AI agent for your business?
You describe the job in plain words. We handle the technical part, test the agent on your real work, and only let it act on its own once it gets the job right.
-
30-minute scoping call
Tell us the job you want off your plate. No technical knowledge needed.
-
Intake form
A short form tailored to your tools, so we know what you use and how the job runs today.
-
Discovery call
With the people who do the job. We walk through real examples and the decisions they make along the way.
-
Proposal with a fixed price
What the agent will do, which tools it works in, where it needs your sign-off, and the spending limit per run.
-
Build
We build the agent and test it on your real examples.
-
Shadow run
The agent drafts and your team decides. It only acts on its own once its work matches your team's.
-
Hand over
It runs in your own account, and one person on your team is trained to manage it.
What do you get with a custom AI agent?
An agent that takes one repetitive job off your plate, works in the tools you already use, sounds like your team, and never acts on anything important without your sign-off.
-
01
It does the work
-
One job off your plate, end to end.
The repetitive work you still do by hand: invoices, follow-ups, ticket triage, reporting.
-
Works in the tools you already use
CRM, support desk, billing, projects and accounting.
-
Sounds like your team and remembers the conversation
A customer who replies weeks later picks up where they left off.
-
-
02
You stay in control
-
Nothing important happens without your sign-off.
You choose what needs approval and which data it can touch.
-
No surprise bills.
Every run has a spending limit, and platform costs go on your own account with no markup.
-
A full record of what it did, when and why
Plus an off switch.
-
-
03
It's built to last
-
Tested on your real work before it goes live
And re-tested when the platform changes.
-
Yours to keep.
It runs in your account, and the code lives in your repository.
-
Ongoing support subscription (optional)
If you'd rather not manage it yourself.
-
Is a custom AI agent right for your business?
Yes, if a repetitive job eats hours every week and someone can explain how it's done. No, if a chat assistant already handles it when you ask.
It's a fit if
- You know what you want it to do, but not how to build it.
- A repetitive job eats hours of your team's week.
- The job means jumping between several tools.
- You tried building it yourself, and it never felt solid enough to go live.
- Nothing important should happen without someone's sign-off.
It's not a fit if
-
A chat assistant already does it when you ask.
You may only need better connections to your tools. See Custom MCP Builds →
-
Nobody can explain how the job is done or what good work looks like.
-
Your tools have no API, so there's nothing for the agent to work in.
Custom AI agents: questions business owners ask
01 / 10
An assistant that does a job for you instead of just answering questions. It reads, decides and acts in your tools, within limits you set, and checks with you when it should.
02 / 10
No. You tell us what the job is and what good looks like. We handle the build and train one person on your team to manage it.
03 / 10
A service that runs AI agents for you, so there are no servers to manage. We build on Claude Managed Agents from Anthropic, launched in April 2026 and still in beta, and on the OpenAI Agents API. The platform runs the servers, and we build the agent and connect it to your tools.
04 / 10
That's what the shadow run is for. The agent works alongside your team on real cases and only acts on its own once its work matches theirs.
05 / 10
You decide what needs your sign-off, which data it can see, and how much each run can spend. Every action is logged, and you can switch it off at any time.
06 / 10
Yes. We build its instructions from your real messages and examples, so its replies follow your tone and your process.
07 / 10
The agent runs in your own account with Anthropic or OpenAI, under your agreement with them. If your data can't leave your control, we run it on Vercel, using only AI models that keep none of your data, or on your own cloud.
08 / 10
You pay the platform directly, with no markup: usage on either platform, plus $0.08 per active hour on Claude Managed Agents. Every run has a spending limit.
09 / 10
You get a fixed price in the proposal, based on the job, the number of tools and the sign-off points.
10 / 10
One person on your team, trained at hand-over. Or choose the support subscription, and we keep it tested and working when the platform changes.
What job do you want off your plate?
Book a 30-minute scoping call and bring one repetitive job. If a simpler tool can do it, we'll tell you. If it needs an agent, we'll show you how we'd build one you can trust.