SW·002 - What we do
We build with you.
You keep the system.
We partner with organisations to design and ship software, and to put AI into the work itself, not as a chatbot on the side. The platform underneath is open to anyone with things to run - at work, in study or at home. SynthWise is the scaffolding those projects run on: assistants, agents and custom models, managed on infrastructure you control, with a record of every call.
How the work is structured
Delivery and platform
are the same system.
We sit with you, map the process as it actually runs, and build it. That is the revenue work: agentic automations, data lakes, products. You leave with running infrastructure you can understand and extend, not a slide deck and a dependency.
Every engagement is built on SynthWise, and every engagement makes SynthWise better. We are our own first customer. Internal operations, client delivery and the product are one loop, not three businesses.
01 - ADOPT
Private AI
Regulated SMEs that cannot send customer data into someone else's cloud and hope - and anyone who would rather their own documents stayed their own.
Assistants, agents and custom models, managed and deployed on your own hardware, sandboxed and ring-fenced.
The barrier is not the models. The models work. The barrier is trust, permissions and a place to run the work without the data leaving your infrastructure. We map the processes you depend on, give each role an assistant with its own tools and bounds, and run them here.
Self-hosted inference sits on the same substrate as the rest of the platform. A law firm does not want to manage vLLM. They want a private assistant that processes documents without the files leaving the building. We deliver the outcome, not just the GPU. And it is not all or nothing. Model connections are workspace assets, so the workspace holding client files runs on the private model and nothing else, while the marketing workspace next door keeps the frontier models and the image and video generators. Mix and match, per workspace, on one platform.
- Custom LLMs on hardware you control, classified INTERNAL by default
- Model access set per workspace: frontier and image models here, the private model only there
- Assistants with instructions, tools, project access and an audit trail
- Give an assistant an automation and it acts as an agent, still inside the same guardrails
- Unknown providers classified EXTERNAL, because that is the safe assumption
02 - BUILD
Software
Credit unions, regulated SMEs, and anyone whose reporting still lives in vendor exports - or who just needs a small tool built well.
Enterprise software that works with you and for you. Data lakes, warehouses, reporting, integrations and products.
The governed data lake we built for a UK credit union is the example. Vendor exports in, a lake and warehouse underneath, live operational reporting out: arrears, exposure, member activity, under the audit and data-protection obligations that organisation actually carries. The connectors, schemas and trail are reusable. For the next credit union this is configuration, not a ground-up project.
We design and ship through our own Studios. The tooling is why it moves at a speed that was not previously available to an organisation this size. The software can sit under SynthWise or stand on its own.
- Governed lake and warehouse, built once for the sector
- Live reporting under the obligations you already have
- Integrations into the systems you already run
- Products and sites when that is the right shape, not a platform you did not ask for
- Small builds too: a page, a tool, a prototype for a course or a household
03 - RUN
Platform
Anyone ready to run their own AI operations - a team, a student, a household - with a UI that can shrink as the work gets simpler.
A place to manage and deploy assistants and agents while software interfaces are still catching up.
Over time, more of the interface will be natural language. Features of the old UX will disappear. In the meantime you need scaffolding: a platform on which to manage agents and to build the applications that replace the spreadsheet. SynthWise fills that gap, and is engineered so natural-language features can replace traditional ones without throwing the system away.
Data and its provenance get more valuable, not less. The APIs you already pay for stay wired in. You keep the context, the permissions and the record when a better model arrives.
- Five layers, one substrate, one permission model
- The same substrate for a business, a course of study or a household
- LLM-agnostic: swap models without rebuilding the work
- Export is standard. Leaving is a button, not a negotiation
- Built on Rails 8 and PostgreSQL. You can read the architecture.
We love our own cooking
We run the business
on the thing we sell.
It's not a slogan, it's honestly how the work happens. We do our prototyping, design, and planning inside the platform, so most features exist because we needed them ourselves first. When something is awkward, we're the ones who feel it, and it gets fixed. Client feedback and our own experience evolve the platform together.
The stack is deliberately plain. A large Ruby on Rails monolith, kept simple on purpose, partly because Ruby is token-efficient for LLMs, and mostly because a simple, single codebase is one an AI can actually reason about. Fewer moving parts, fewer places to get lost.
Building this way is the most creative and productive we've ever been making software. Things that weren't possible for a team our size are now an ordinary afternoon. The platform is how we try to pass that on - to people who don't write code, for whatever they have to run.
Next step
Bring the process
everybody dreads.
We map it, run it on SynthWise, and hand you the receipts. If it works, the next process is mostly configuration.