AI development and engineering
AI engineering services for the product you sell or run
We add AI features to the products you sell: assistants, voice interfaces and agents. Our engineers also run the infrastructure such features depend on, so we connect them to the systems you already have, build the model serving and retrieval behind them and test them before your customers see them.
Our team has built software and production AI for
Four layers, planned together
What an AI feature costs per request, how quickly it answers and how good its answers are all depend on the model, the retrieval and the infrastructure beneath it. We plan the four layers together, so their trade-offs are settled in one design instead of between separate suppliers.
Agents in your product
Custom agents, MCP servers and A2A. A design uses several agents only where the work divides cleanly.
Applications
Assistants and voice interfaces, and applications that answer from your own documents.
Models
Choosing a model, fine-tuning it where that pays off and testing it on the task it has to do.
Infrastructure
GPUs, inference servers and cloud setup, in your own account or with a provider in the EU.
Where cost and failures start
Cost and failures start where two layers meet: a model larger than its task needs, retrieval that has gone stale, or tool permissions wider than the feature requires. We review those joins first.
In the product
Access to your systems
Underneath
Systems we have run in production
Built and operated by our engineers.
Models
Computer vision models trained on over 100 million images
Applications
Voice AI with responses in under 600 milliseconds
Agent systems
Enterprise assistants that combine retrieval, agents and MCP
Infrastructure
Cloud platforms behind public apps used by hundreds of thousands of people
Project details are available on request.
Talk to an engineerAI integration services for the product you already run
An AI feature becomes useful once it reaches your own systems. Most integrations follow the same path: read the records the user is authorised to see, draft or perform one bounded action, and return the outcome to the workflow where the request started.
Consider a hypothetical support ticket. The assistant reads the ticket, the customer's order history and relevant help articles using read-only credentials. It drafts a reply proposing a refund, and a support agent approves it before anything is sent. Issuing the refund is a separate write permission, added later only if the team wants it.
- Where your existing API or SDK does the job, we call it directly. An MCP server is worth adding when several AI clients need the same tools; for a single feature it is often unnecessary.
- Each action is scoped to specific records and fields, with a named approver where the risk calls for one.
- Failed and repeated calls are handled explicitly. We define which steps are safe to retry and use idempotency keys or status checks so a timeout does not trigger a second refund. That lowers the risk without removing it.
What we hand over, and when it counts as ready
You receive a working integration your team can run and change. It includes the agreed data flow, the permissions it uses and the integration code. Tests cover realistic failures like expired tokens or a slow upstream service. We report measured response times and cost per request, and write deployment instructions for your environment.
Acceptance is agreed up front, with a named owner on your side who signs off. Where the workflow needs caution, the first release can stay read-only or require approval for each action.
Tested on requests from your product
Each feature gets a test set built from real requests. We track the cost and the response time of each request, and the quality of the answers, before and after every change.
Agents for business processes, such as claims handling or invoice checks, have their own page.
How we handle access
We take no money from model or tool vendors, so what we recommend is what tested best for you.
The agents we build run in your environment.
Each tool is limited to the access its task requires.
Every agent action goes into a log.
Where you have a works council, we work through its requirements before launch.
Questions
What are AI engineering services?
In AI engineering services, engineers design, build and run the AI features of a software product, along with the models, retrieval and infrastructure behind them. At Cloudsail we plan four layers in one design: agents, applications, models and infrastructure. We connect each feature to the systems you already have and measure its cost, response time and answer quality on a test set built from real requests. You receive a working integration your team can run and change.
What does a first conversation involve?
Thirty minutes with one of our engineers. It is free, and we need no access to your systems or data for it.
What kind of AI development company are you?
An engineering company. The engineers who advise you also build the feature in your product and your cloud, test it on requests from your product and hand over the code, test sets and runbooks to your team. We don't resell tools, and no model vendor pays us.
Can you integrate AI into the systems we already run?
Yes. We connect models to your existing APIs, databases and business systems through tools or MCP servers, with each tool limited to the access its task needs. We start from your current architecture rather than a new platform.
How do you take an AI pilot to production?
We first build a test set from real requests and measure the pilot's cost per request, response time and answer quality against it. The feature goes live with the monitoring, access limits and fallback it needs, and the same tests run before every later change.
Why does it matter that you also handle models and infrastructure?
Because cost and limits are decided at those layers. Should a feature call for a model hosted in your EU account, or for a small model trained on one high-volume task, that work becomes part of the same project rather than a separate one.
Do you also automate business processes?
Yes. Agents for complex cases in your operations, such as claims or invoice checks, are covered on their own page.
Which frameworks do you build with?
The ones your team will maintain after we leave, such as LangGraph, the OpenAI Agents SDK, the Claude Agent SDK, Google's Agent Development Kit or Mastra. Where the control flow is simple, we write plain code instead.
When is more than one agent worth it?
When separate parts of the work belong to different owners, run on different tools or come from different vendors. Each hand-off adds latency, cost and one more thing that can fail, so we start with a single agent and split only where such a boundary really exists. A2A is useful at exactly those points.
Are you tied to one vendor's platform?
No, and no vendor pays us. Your current models, clouds and frameworks are the starting point, and we suggest alternatives only after they have done better on your tasks.
Can you work with our current model and provider instead of replacing them?
Yes. We start from your current architecture and test the model you already use on real requests. If it meets your quality, latency and cost targets, it stays. If it falls short, we show the measured trade-offs against alternatives and you decide whether switching is worth it.
What determines the scope and price of AI integration?
Four things: the systems involved and how we get access, the risk of the actions the AI may take, the evaluation work needed, and the operating work after launch. We agree scope after a call with an engineer. The first 30-minute conversation is free.
Related services
AI agent evals
Find out how often an agent gets the job done before it acts in your systems, by checking outcomes, tool calls and how reliably runs repeat.
AI automation
Routine work taken off your teams: processes automated step by step with rules, workflow tools, models and agents, approved by people and measured per case.
AI chatbots
Customers and staff answered faster from your data, with approved actions and a handover to people when it matters.
AI strategy and rollout
AI strategy and rollout that starts with one running, measured system on your own processes and builds the roadmap around its figures.
Data integration for AI
Answers from file shares, SharePoint, ERP and tickets in seconds, for assistants and agents, with permissions carried through and tested.
AI engineers for your team
Extra AI engineering capacity when your team lacks AI skills: experienced engineers who build with you and leave the know-how.
AI evals engineering
Evals built from your own traces show whether each change made your AI product better or worse, with a release gate that stays in your repository.
AI system testing
Before go-live, acceptance tests on your own cases tell you how often your chatbot, assistant or agent is right, in terms your QA team already uses.
Free AI usage review
One page for management on where your AI money goes, whether it pays off and what to change, from exports you choose to share. Free, no system access.
Enterprise AI
One approved AI platform for the whole company: an internal assistant instead of shadow AI, approved models, data that stays in-house and spend per team.
LiteLLM gateway
Budgets and access rules for every team's AI use, through a LiteLLM gateway that is hardened, connected to your identity provider and kept up to date.
AI platform for all teams
One approved route to AI for every team, with budgets, logs and a free choice of model, run on your servers or in your cloud.
System access for agents (MCP)
AI agents that look up and update data in your systems, with narrow permissions, human approval for changes and tests.
On-premise AI
AI on offers, contracts and customer files, with the data staying in your company: own servers or own cloud account, tested before go-live.
Azure OpenAI and Bedrock
GPT and Claude on company data, under the cloud contracts you already have: Azure OpenAI and Amazon Bedrock in your own account, with EU processing and one gateway.
Knowledge assistants (RAG)
Staff find answers in internal documents in seconds, with sources cited, permissions kept and tests on your own questions.
RAG evaluation
An assistant that answers from your documents, tested on retrieval and on answers in each language your users write in, so you know how often it is right and cites the right source.
vLLM on Kubernetes
Open-weight models on your own GPUs, so data stays in house and costs follow capacity you plan, sized from measured load.
From the blog
AI pilot to production
What has to be in place before a successful AI pilot goes live: regression tests, versioning and rollback, monitoring, an owner, a handover and a staged rollout.
LLM latency budget
How to find out why an AI feature is slow: split each request into queueing, retrieval, model and tool spans, then fix the largest one you can change and measure the tail under load.
Self-hosted AI or managed API
How to decide whether an AI workload should run on your own infrastructure or stay on a managed model endpoint, using access requirements, operating capacity and measured costs.
Talk to an engineer
Bring one piece of work you would like to improve. The call takes 30 minutes and needs no access to your systems or your data.
Booking runs on Google Calendar.

Mickey Graf
Founder and CEO of Cloudsail
Prefer to write first? Leave your email and a few lines. Mickey reads every message himself and replies by email with times for a call.
Look for the draft in your email program
Send it from there, and Mickey replies in person with times for a 30-minute call. If no draft has opened, write to us at miki@cloudsail.com.