Brand Voice Setup · £350
Your tone, baked into AI.
30-minute intake, one-page voice document, 12-month reference catalogue.
Read the breakdownApproach
How I work, and what I work with.
A clear four-step process and the architecture underneath it: open-source and open-weight models where they fit, frontier models where the stakes justify them, and Hermes for orchestration. The same approach on every project.
Four steps, no surprises.
01
Listen to your business, not just your tech stack.
The first conversation is about the business: who you sell to, what costs you time, what protects revenue, what your team actually does day to day. AI only matters where it lands on those answers.
02
Decide honestly on what to build (or not build).
You get a clear recommendation on what is worth automating, what isn't, and what isn't ready yet. Sometimes the right answer is "not yet" or "use this off-the-shelf tool". Sometimes it's a custom build. The recommendation is the same either way.
03
Build it end to end, with fixed scope and price.
If the answer is "build", scope, price, and timeline are agreed up front. Then I ship. A six to eight week engagement is typical, with a working production system at the end and ownership handed over.
04
Hand over, and stay accountable.
You own the code, the data, the documentation, and the roadmap. You get training for your team and thirty days of post-launch support. The system stays maintainable long after I'm gone.
The stack: open where it works, frontier where it matters.
Maintainable by design
Hermes Agent · orchestration
The open-source agent system that runs underneath every client project. It plans the work, monitors what is running, gives you a single place to see and audit what is happening, and routes each task to the right model. When a model improves or a task changes, the swap is a config change, not a rewrite.
Open sourceSelf-hosted
Open source and open weight by default
Right-sized models for routine work
Open-source and open-weight models (minimax, DeepSeek, GLM) handle classification, extraction, drafting, triage, the everyday work. They are picked because they fit the task, and because the architecture is set up to swap in stronger models when a workload demands it, without rewriting the workflow.
minimaxDeepSeekGLM
Frontier where stakes justify
Frontier models for the moments that matter
For the security audit, the final draft review, the high-stakes ten percent where the cost of getting it wrong is bigger than the cost of a more capable model. Claude and GPT, called only when the workload warrants it.
ClaudeGPT
Why the architecture matters more than the model choice.
Most AI projects pick one model and use it for everything. Either a frontier model on every task (expensive, slow, often wasteful) or an open-weight model for everything (cheap, but brittle when the workload gets serious). Both approaches leave money on the table and quality behind.
The right architecture routes the task to the model that fits it. High-volume work goes to open-source and open-weight models. The security audit, the final review, the moments where getting it wrong is the more expensive outcome go to frontier models. Hermes handles the routing automatically, so the cost discipline comes from the system, not from cutting corners on the work that matters.
Monitoring, fallback paths, and clear upgrade routes are built in, not bolted on. Reliability comes from the discipline of the architecture, not from any single model being good enough.
Or skip the full build
Fixed-scope products with a clear handover.
Start with your brand voice, local search presence, website or the full service-business launch. Each product has a defined price, timeline and result, with no platform lock-in.
Local SEO · £149 audit
Find out why your competitors are in the Map Pack and you aren't.
For UK trades businesses. One-off audit: GBP score, Map Pack position, citation health, website local signals, competitor comparison. 1-page report, plain English, 5 working days.
Read the breakdownWebsite Build · From £300 setup
Bespoke site, .co.uk, email.
All-in-one bundle: build, domain, mailboxes, hosting, updates. Three tiers.
Read the breakdownMercury OS · From £1,495 setup
A working AI business, in a box.
Four production modules installed on your own hardware. Hermes Agent, Vereby, S0cial Master and the Zoho Lead Engine. Founder pricing for the first cohort.
Read the breakdownIf any of this looks like a workflow you have, twenty minutes is enough to see if I can help.
Tell me which workflow looks closest to a problem you're sitting with right now. I'll tell you honestly whether AI is worth building it, and what the first useful step looks like.