Services
Support & Maintenance
Managed hosting, monitoring, updates and security patching after launch.
Software degrades without care — dependencies age, threats evolve, performance drifts. Launch is the midpoint, not the finish line.
For AI features specifically, that decay shows up differently than a normal outage. A model's answers can quietly get worse as the real-world data it's reasoning about shifts, well before any dashboard turns red — exactly the kind of drift a standard uptime monitor was never built to catch.
Our managed support keeps everything we build — and things we didn’t — secure, fast, and improving, under plans sized to how much change you expect. Framework and dependency patching runs on a defined cadence, not on discovery. For AI features specifically, that means monitoring so you know when a model’s answers are drifting, not just when the server is down, and an incident process for exactly what happens when something breaks — the three things a prototype almost never has and production always needs.
Drift detection only has value if it triggers action, so ours is wired to do something, not just report it — repeated drift alerts kick off a review of the affected feature, and where the fix is a retrained or adjusted model, we validate it against real production traffic in shadow before it ever replaces what's live. The same defined-cadence discipline that covers dependency and security patching for the rest of the platform covers this too: it's a scheduled practice, not a fire drill.
Frequently asked
What does it take to move an AI prototype into production?
Usually three things a prototype doesn't have yet: monitoring so you know when it's wrong, a maintenance plan for the models and dependencies it relies on, and an incident process for when something breaks. We fold all three into support plans rather than treating them as a separate project.
What's covered in a monthly maintenance report?
Uptime, security patches applied, dependency updates, performance metrics, and a plain-language summary of what changed and what we recommend next.
Do you offer different support tiers?
Yes — Essential covers security and stability, Growth adds content and optimisation, and Premium adds priority response and quarterly strategy reviews. We size the plan to how much change you expect on the platform.
What actually happens when model drift is detected?
It doesn't just get logged. A repeated drift signal triggers a review of the affected feature, and if the fix is a retrained or adjusted model, we validate it against real production traffic in a shadow deployment before it replaces what's live — the same care as a code deploy, just applied to a model instead.