AI implementation · RAG · Claude Skills & MCP · Cloud
We build AI into your systems — and your next product.
RAG that answers from your own documents and data. Claude Skills and plugins that give AI your team’s tools and know-how. Agents that finish the job. Whole AI products, from idea to launch. Built by engineers who run AI in their own live products, on cloud we design ourselves.
What we build
AI implementation, from first use case to production
Not a demo that works once. Systems your team relies on, with the guardrails, cost controls and handover that production needs.
RAG & knowledge systems
Answers from your own documents and data, with citations a reviewer can open and check. We design access to follow the source system and choose retrieval to fit the corpus. When the sources do not support an answer, it says so instead of guessing.
Shipped: the policy assistant in AetherERP, our AI-native ERP, answers HR questions from the handbook and discards any citation it did not actually retrieve.
Skills, plugins & MCP integrations
Your recurring workflows packaged as Claude Skills, so the assistant does them the same way every time. Plugins and MCP servers that connect it to your tickets, databases and internal APIs — with only the permissions each task needs.
The groundwork, shipped: Stock Tracker’s assistant answers portfolio questions by calling eleven tools against live data, capped at six steps. MCP servers expose tools through that same mechanism; Skills add the instructions around them.
Agents & workflow automation
Agents that finish a task rather than start one: bounded loops, a human review queue when confidence is low, token budgets that stop runaway spend, and a log of every model call.
Shipped: the App Generator’s fix agent reads real CI failures and patches code, within a set number of attempts, until the build passes — every run, failures included, is public on the benchmark page. AetherERP’s AI gateway routes each task to a model and sends low-confidence results to a person.
AI products, built and run
Got an AI product idea? We take it from prototype to a live product — the application, the AI features inside it, and the cloud it runs on. We know the road because we build and run our own.
Shipped: Tally, Stock Tracker, AetherERP and every tool on this site — all live.
What a free review sends back · illustrative
“Invoice intake: AI reads each invoice and suggests the matching PO; plain rules check totals and tolerances; a person approves anything under 90% confidence. About four weeks.”
And the ground it all stands on — architecture, Terraform, security review and cloud cost on AWS, Azure and GCP. All services →
Track record
Cloud at scale, before AI was in the brief
Our cloud and infrastructure judgement comes from delivering, directly and through partners, for teams at:
Prime Therapeutics
Pharmacy benefits
UHG Optum
Healthcare
Change Healthcare
Healthcare data
Carnival Cruise
Travel & hospitality
Raytheon
Aerospace & defense
FCMAT / CSIS
Public sector
Engagements delivered directly and through contract and subcontract partners. Company names identify where the work was done and do not imply endorsement of these products.
Try it yourself
Free AI tools you can use right now
Built for our own engagements, running on this site. No account, no waitlist, no demo request.
App Generator
Describe the application. Get architecture analysis, a tech-stack recommendation, UI/UX and SEO guidance, a wire diagram, and a downloadable scaffold — built on real GitHub Actions and fixed until it passes.
Open the wizard →IaC Review
Paste the Terraform you already have. Get an opinionated read on security posture, availability, and the choices that will page someone at 2am.
Review my IaC →IaC Generator
Infrastructure as Code from a plain-language description of what you need running — across AWS, Azure and GCP.
Generate infrastructure →Also running here: Code Review · Resource Optimization · Review History
What makes the App Generator different
We don’t hand you code and wish you luck
Most generators stop at plausible-looking files. Ours keeps going until the build is actually green, on real CI, before anything reaches you.
Detect the stack
The generated project is read to work out what it actually is, and a matching CI workflow is injected.
Run it for real
It is pushed to a throwaway private repository and built on GitHub Actions — the same CI your team uses, not a simulation.
Read the failures
Jobs, annotations and logs are read back, and a fix agent works the errors the way an engineer would.
Green, then clean up
It re-runs until the build passes or guardrails stop it, then the throwaway repository is deleted.
You get the result either way — including an honest report when a build could not be made to pass.
Also from us
Products we build and run
Same platform, same deployment discipline, same review tooling as everything above.
Tally
Time and expense capture that does not need chasing
Consultancies, contractors and any team that bills for its time
Read more →Stock Tracker
Market and IPO monitoring that tells you when something changes
Investors and analysts who would rather be notified than go looking
Read more →AetherERP
An ERP built AI-native instead of AI-retrofitted
Operations and finance teams outgrowing spreadsheets and bolt-ons
Read more →Common questions
Before you try it
Can you build a RAG system over our own documents and data?
+
Yes. It answers from the sources you choose, with citations a reviewer can check, and says so when the sources do not cover a question. We design access so each person sees only what the source system already allows. Retrieval fits the corpus — ranked keyword search, or embeddings and a vector index — and by default it runs in your own cloud account.
What are Skills and plugins, and why would we want them?
+
A Skill packages one of your recurring workflows — the steps, the templates, the checks — so an AI assistant does it the same way every time instead of improvising. Plugins and MCP servers connect the assistant to your systems: your ticketing, your database, your internal APIs, with only the permissions each task needs. Together they turn a general chat model into something that does your team's actual work.
We have engineers. Why bring you in?
+
To get the first production system right, faster. Retrieval quality, tool permissions, cost control and what happens when a model is slow or wrong are problems we have already worked through in our own products, such as low-confidence results routed to a person and citations checked against what was actually retrieved. We work in your repository alongside your team, and hand over a system they can run and extend without us.
What does it cost?
+
There is no price list, because no two scopes are the same. The first review is free. Fixed-scope work is quoted before it starts, so the number does not move. When a job is small, we often just do it.
Does our data go to a model provider?
+
Only to the provider you approve, and we keep each request to what the task needs. By default we build into your own cloud account, so your documents and indexes stay in your storage. Model calls go to the provider's API under its commercial terms; Anthropic's commercial terms, for example, say API data is not used to train its models.
Which AI models do you work with?
+
Mostly Anthropic's Claude, and OpenAI-compatible models where they fit better. In AetherERP each task is routed to a model through configuration, with prompt caching and a token budget, and we build the same way for clients: changing model is a setting rather than a rewrite, and cost per request is visible from day one.
Do you only sell tools, or do you do the work?
+
We do the work. Our focus is AI implementation — RAG systems, Skills, plugins, agents and AI products — along with the architecture and cloud infrastructure around them. The tools on this site are the parts of that work we automated for ourselves, and they are free to use.
What else have you built?
+
Tally for time and expense capture, Stock Tracker for market and IPO monitoring, and AetherERP, an AI-native ERP built on .NET and React. All three are live. They run on the same platform as these tools, and they are where the patterns on this page — citations, tool limits, human review, token budgets — were proven first.
What does the App Generator actually give me?
+
An architecture analysis, a tech-stack recommendation, UI/UX and SEO guidance, a wire diagram and a downloadable project scaffold — with an infrastructure cost estimate that updates as you choose options.
How do you know the generated code works?
+
We run it. The project is pushed to a throwaway private repository and built on real GitHub Actions; failures are read back from the jobs, annotations and logs, a fix agent works them, and it re-runs until the build is green. Then the repository is deleted. If a build cannot be made to pass, you get told that rather than handed code that does not compile.
Ask where AI fits — it costs nothing
Tell us the process you want AI to take on, or send the architecture or Terraform you are unsure about. Within a few days you get a short written plan: what we would build or change, what it would take, and what to avoid. No call required. Or try the App Generator and watch it prove its code on real CI.
We reply within one business day.