How we work

We are paid to compress time, so we work in short, measured steps, keep you free to change models and vendors, and put everything we build in your name.

Provider-agnostic by design

A thin model layer sits between your agents and the model providers. OpenAI, Anthropic, xAI Grok, Google, Mistral or open-weight models can be swapped per task without touching business logic.

Evaluations decide, not opinions

We build a test set from your real cases and score each candidate model on quality, latency and cost. The numbers choose the model.

Platform chosen per client

Azure, Cloudflare or your own infrastructure, decided by where your data must live, the tools your team already uses and the managed services you need.

Right-sized and cost-controlled

Small, fast models handle routing and extraction; frontier models handle hard reasoning. Cloud and AI usage have budgets, per-user limits and alerts.

The engagement, step by step

  1. Baseline

    Two weeks measuring the current situation together: hours spent, errors, delays, costs. It is the reference every result is compared against.

  2. Initial package

    The highest-value scope first, delivered in about three months for a fixed price. You pay in monthly instalments, each one only after you accept that month’s delivery.

  3. Warranty

    30 days after acceptance during which any defect is fixed at no cost.

  4. Continuity service

    A monthly fee covering monitoring and alerts, incident fixes, security updates, channel API changes, tested backups, cost control and a one-page monthly report.

  5. Further modules

    Each one priced separately, in order of risk, once the previous step has proven its value.

Response times in the continuity service

SeverityExampleResponse
CriticalA core sync or AI access is down4 business hours
ImportantOne channel or automated process failsNext business day
MinorIsolated errors with no impact on operations5 business days

Commitments in every project

  • Code, cloud accounts and data in your name from day one.
  • Cloud and AI usage billed to you directly by the provider, with an estimate before we start.
  • Any AI action that changes data needs approval from an authorised person.
  • Documentation and a second developer reviewing code, so nothing depends on one person.
  • A short monthly meeting and a formal review of each delivery before it is invoiced.

Bring one slow process. Leave with a plan to compress it.

A 90-minute working session with your team. No slides.

Book a working session