AI readiness review
A short, honest assessment of where AI will and won't pay off in your business, with a prioritised list of use cases.
AI that runs where you need it: in the cloud, on your own hardware, and in your team's hands.
Cirrumuse is a small UK consultancy. We build cloud platforms and agentic software, install private language models that never leave your network, and train your developers to work AI-first.
- UK based
- Fixed price or time and materials
- Your data stays yours
- Plain-English communication
What we do
Six services that fit together. Most clients begin with one and add another once it has paid off.
-
AI solutions
Practical applications of modern language and machine-learning models, applied where they genuinely help: document, email and data processing, knowledge assistants over your own content, workflow automation and decision support.
-
Cloud solutions
Reliable, secure and cost-aware infrastructure with a strong focus on Amazon Web Services: architecture and migration, infrastructure as code and CI/CD, serverless and containers, security reviews and cost optimisation.
-
Agentic development
Software agents that plan, use tools and finish multi-step tasks, with humans firmly in control: custom and multi-agent workflows, MCP connectors, approval steps and full audit trails.
-
Local & private AI
Language models running on hardware you control, so sensitive data never leaves your network. Details below.
-
AI-first team training
Hands-on workshops that make your developers faster and safer with AI coding tools. Details below.
-
Legacy modernisation
Move ageing systems onto current platforms without stopping the business, from lift-and-shift to full rewrites on modern architecture, using AI to understand and test old code. Details below.
Not sure where to start? Start small.
Proof of concept
A working prototype on your own data, a clear result and a written business case. Fixed price.
Local AI starter
A private model installed on your hardware or cloud account, connected to a chat interface and a small set of your documents.
Legacy assessment
A short review of an ageing system: what it does, what it costs to keep, the risks, and a staged route off it.
Team workshop
One or two days of practical AI-assisted development training, using your real code and tools.
Local & private AI
Some data should never be sent to an outside service. Open-weight models now run well on hardware you own or on a private cloud account, and we set them up so they are useful on day one.
Typical reasons clients ask for it: client confidentiality, regulated data, intellectual property, predictable costs, or simply working without an internet connection.
- Hardware and sizing advice. Which model fits which GPU, workstation or server, and what it will cost.
- Installation and tuning. Open-weight models served with tools such as Ollama, vLLM or llama.cpp, quantised for your hardware.
- Private knowledge. Retrieval over your own documents, with access controls that match your existing permissions.
- Everyday interfaces. Chat front ends, editor integrations and APIs your existing tools can call.
- Air-gapped options. Fully offline deployments for sensitive environments.
- Hybrid design. Local models for private work, cloud models for heavy lifting, with clear rules on what goes where.
Train your team to build AI-first
Tools like Claude Code, GitHub Copilot and Cursor change how software is written. Teams that learn the habits early ship faster; teams that don't end up with unreviewed, inconsistent code.
Workshops use your own repositories and standards, run on site or remotely, and leave you with written guidance your team keeps using.
- AI-assisted development. Prompting, planning and working with coding agents on real tasks.
- Review and testing. How to check AI-written code properly, and write tests that catch what models get wrong.
- Team standards. Shared rules, project instructions and guardrails so results are consistent across the team.
- Tool rollout. Choosing, configuring and securing the assistants your organisation will use.
- Leaders' briefing. What AI changes for planning, hiring, risk and cost, without the hype.
- Acceptable-use policy. Practical rules for staff using AI at work, written in plain English.
Legacy modernisation
Old systems often run the business and nobody dares touch them. We help you understand what they do, prove it before changing anything, and then move them to platforms you can maintain, by rehosting, replatforming or rewriting, whichever fits.
We favour small, reversible steps (including the strangler-fig pattern for rewrites), so the old and new systems run side by side until you are happy to switch off the old one. AI speeds up the slow parts: reading undocumented code, recovering how it works and writing the tests that protect it.
- Assessment and roadmap. What you have, what it costs to keep, where the risks are, and a staged plan with honest estimates.
- Code and knowledge recovery. AI-assisted reading of undocumented code, producing plain-English documentation and business rules.
- Safety nets first. Tests that capture current behaviour before anything changes, so you can prove the new system matches.
- Rewriting with modern architecture. Rebuilding applications as clean modular monoliths, microservices, event-driven or serverless systems, with APIs, automated tests, CI/CD and infrastructure as code.
- Incremental migration. Replace one piece at a time instead of one risky cutover, with a rollback plan at every step.
- Platforms and languages. Older .NET, Java, VB6, PHP, COBOL and mainframe workloads, Access and Excel applications, and on-premises servers moving to AWS.
- Data and integrations. Database migration, wrapping old systems with modern APIs, and retiring brittle file transfers.
The full AI stack, from data to production
Whether you need a quick proof of concept or a production platform, we cover every layer.
- Generative AI & LLM applications
- Chatbots, copilots and assistants; retrieval-augmented generation, vector search, prompt management, multimodal documents, images and audio.
- Model fine-tuning
- Supervised fine-tuning, LoRA/PEFT, preference tuning and distillation to smaller, cheaper models. Open-weight and hosted options.
- MLOps & LLMOps
- Training and deployment pipelines, model registry and versioning, data pipelines, monitoring for drift, quality, latency and cost.
- Guardrails, safety & governance
- PII redaction, content policy, prompt-injection defence, grounding checks, audit logs and UK GDPR / EU AI Act alignment.
- Evaluation & red-teaming
- Automated test sets, regression testing across model upgrades, adversarial testing and human review workflows.
- Agent orchestration
- Multi-agent workflows, tool and API integration, Model Context Protocol servers, approval gates, tracing and replay.
- AI cloud infrastructure
- Amazon Bedrock, SageMaker, Azure AI and Vertex AI; GPU and inference optimisation; private VPC deployments; token and GPU cost control.
- Data engineering for AI
- Ingestion, cleaning, labelling, data lakes and warehouses, lineage, synthetic data and access controls.
- Classical ML & analytics
- Forecasting, anomaly detection, classification, computer vision, OCR, speech-to-text and text-to-speech.
- Strategy & support
- Readiness assessments, use-case prioritisation, proofs of concept and ongoing managed support.
A simple, transparent process
No surprises. You know what is being built, what it costs and what happens next at every stage.
Discover
A free initial conversation to understand your goals, constraints and whether AI is even the right tool.
Propose
A written scope, timeline and either a fixed price or a time-and-materials estimate. Nothing starts until you agree in writing.
Build
Short iterations with regular demos, so you see working software early and can steer as we go.
Deliver & support
Documentation, handover and optional ongoing support. You own what we build for you.
Two ways to contract, chosen to fit the work
Fixed price
Best when the scope is clear: a proof of concept, a defined build, a workshop or an assessment. You know the cost before we start, and changes are agreed in writing first.
Time and materials
Best when the scope will evolve: large modernisation programmes, research-heavy AI work or ongoing support. You pay for the time worked, with regular reporting, an agreed budget estimate and the freedom to change direction.
Not sure which suits you? We will recommend one in the first conversation, and large projects can combine both, for example a fixed-price assessment followed by time-and-materials delivery.
Working with us should feel safe
Handing over data and systems requires trust. These are the commitments we make to every client.
Your data stays yours
We only use your data to deliver the agreed work. We never sell it, and we never use client data to train models. Where third-party AI providers are involved, we tell you which ones and what they receive.
Secure by design
Least-privilege access, encrypted data in transit and at rest, secrets kept out of code and multi-factor authentication on the accounts we use to work for you.
UK GDPR aware
We handle personal data in line with UK data protection law and will sign a data processing agreement or NDA before work begins whenever you need one. See our Privacy Policy.
Humans stay in control
Agentic systems we build include approval gates, limits on what an agent may do and logs of every action, so you can always see and reverse what happened.
Clear commercial terms
Written proposals, itemised invoices and plain terms of business. Cirrumuse is not VAT registered, so the price we quote is the price you pay, with no VAT added.
Real people, real contact
A direct phone number and email, answered by the person doing the work. No call centres and no ticket queues.
Where cloud meets muse
The name joins cirrus, the high, light clouds, with muse, the source of inspiration. It reflects what we do: build on the cloud, and bring creative, intelligent AI to the problems that matter to you.
Cirrumuse is an independent UK consultancy. We are deliberately small, so every project gets senior attention, honest advice and no unnecessary overhead. If a simpler, cheaper solution will do the job, we will tell you.
- Company name
- Cirrumuse Ltd
- deankirby@cirrumuse.com
- Telephone
- 07590 217486
- VAT
- Not VAT registered
- Company number
- 17352278
- Registered office
- 3 Avondale Avenue, Worcester Park, London, England, KT4 7PD
Questions people usually ask
How much does a project cost?
It depends on scope. After a free conversation we put a written proposal in front of you, as either a fixed price or time and materials, depending on how well the scope is defined. Small proofs of concept are quick and affordable, and we will advise honestly if a project is not worth doing.
Can a local model really replace a cloud one?
For many everyday tasks, yes: summarising, drafting, searching your documents and answering questions. For the hardest reasoning, cloud models are still ahead. We help you split the work so private tasks stay local and only the rest goes out.
What hardware do we need for local AI?
It depends on the model and how many people use it. A capable workstation GPU is enough for a small team; larger groups need a server or a private cloud instance. We size it before you buy anything.
Should we rewrite our old system or just move it?
It depends on the system. Sometimes moving it to the cloud as it is does the job. When the design itself is holding you back, we rewrite it on modern architecture. Either way we usually do it piece by piece, with the old and new running together, and we will tell you honestly which route is worth the cost.
Will you see our confidential data?
Only what is needed for the work, and only with your permission. We are happy to sign an NDA first and can often work with anonymised or sample data.
Who owns what you build?
You do. Unless agreed otherwise in writing, intellectual property in bespoke deliverables transfers to you on payment. See the terms for detail.
Do you only work with AWS?
AWS is our main platform, and this site itself runs on it. We can also work with Azure and Google Cloud where there is good reason to.
Can AI agents be trusted to act on their own?
Only within limits you set. We design agents with narrow permissions, approval steps for sensitive actions, and monitoring, so autonomy is earned rather than assumed.
Do you offer ongoing support?
Yes. Support and maintenance can be agreed as a monthly arrangement, or you can take over the system entirely using the documentation we provide.
Where are you based and how do you work?
We are UK based and work with clients remotely across the UK and beyond. In-person meetings and on-site workshops can be arranged where it makes sense.
Let's talk about what you need
The first conversation is free and carries no obligation. Tell us what you are trying to achieve and we will give you an honest view.
We aim to reply within one working day.