Everything is on the record
Every prompt, tool call, change and approval goes into an audit log that can only be added to. Practitioners can read it. Only the platform can write to it.
A whole digital team on day one. Every decision still yours.
Civicwright gives your service a multidisciplinary team of AI practitioners, from user researchers to security architects. They work to the Service Standard, challenge each other's work, and stop for a named person at every decision that matters.
Try it below: you're the service owner.
Team
- Senior product managerFraming the problem
- Senior user researcherPlanning research
- Service designerMapping the journey
- Lead technical architectCosting options
- Service ownerDecisions come to you
Activity
Senior product manager drafted the problem statement from the policy paper.
Lead technical architect challenged it: it assumes an online-only service. Point 3
Senior product manager reworked it to cover phone and paper applications. Point 3
Service Standard evidence
- Point 1, Understand users and their needs: no evidence yet
- Point 2, Solve a whole problem for users: no evidence yet
- Point 3, Provide a joined up experience across all channels: no evidence yet
- Point 4, Make the service simple to use: no evidence yet
- Point 5, Make sure everyone can use the service: no evidence yet
- Point 6, Have a multidisciplinary team: no evidence yet
- Point 7, Use agile ways of working: no evidence yet
- Point 8, Iterate and improve frequently: no evidence yet
- Point 9, Create a secure service which protects users' privacy: no evidence yet
- Point 10, Define what success looks like and publish performance data: no evidence yet
- Point 11, Choose the right tools and technology: no evidence yet
- Point 12, Make new source code open: no evidence yet
- Point 13, Use and contribute to open standards, common components and patterns: no evidence yet
- Point 14, Operate a reliable service: no evidence yet
Spent this phase
£0.00 of £120
Your decision
Modelled on all 53 roles in the Government Digital and Data Profession Capability Framework.
Measured against all 14 points of the Service Standard, every day, not just before assessment.
Designed around the 10 principles of the AI Playbook for the UK Government.
Every service deserves a full team. Few get one.
Government has committed to digital and data professionals making up 10% of its workforce by the end of the decade. In March 2026 they were 6.3%.
So service teams start short: no content designer, a shared architect, research squeezed into the gaps. Civicwright gives every team the full set of professions from the first day, so your people spend their time on the work only people can do.
Source: PublicTechnology, August 2026, reporting Cabinet Office civil service statistics.
What a sprint looks like
You set the problem and the budget. The team does the rest of the week, and you spend Friday deciding.
Monday
You set the problem
Upload the policy paper, set a budget and a deadline. Civicwright staffs a discovery team in minutes.
Tuesday
The team gets to work
Desk research, a journey map and a research plan, each challenged by at least two other professions before lunch.
Wednesday
Your researchers meet users
People run the sessions. Practitioners turn the notes into findings the same evening.
Thursday
Weak spots surface early
A mock assessment panel grills the evidence, point by point, while there's still time to fix it.
Friday
You decide
One page of decisions, each with its evidence, cost and any disagreement. Approve, redirect or stop.
You choose how far it goes on its own
Turn the dial to see what practitioners do by themselves and what waits for you. You set it per project and per phase, and some decisions are never automatic.
Act and tell. Practitioners act, tell you, and you can undo it.
| Action | What happens |
|---|---|
| Draft documents, designs and code | Automatic |
| Critique and rework each other's drafts | Automatic |
| Merge code to the main branch | Automatic, you're told |
| Change scope or backlog priorities | Asks you |
| Deploy to a test environment | Asks you |
| Spend beyond the budget | Always yours |
| Touch real user data | Always yours |
| Move to the next phase | Always yours |
Built to be scrutinised
Public services need more than good output. They need to show their working.
Stop it at any level
Pause one practitioner, freeze a project with a snapshot, or stop everything at once. Automatic breakers trip on runaway spend, loops and anything reaching outside its sandbox.
Your data stays yours
Civicwright runs on your own infrastructure. Each task goes to the right model for the job: Claude through the Anthropic API, or local models through Ollama. Anything sensitive never leaves your hardware.
No surprise bills
Budgets per task, per practitioner and per project. Every task shows its forecast cost before it runs, and hard limits stop it rather than warn.
Practitioners that get better, and show you how
Each practitioner is built from a real role level in the Capability Framework, with that role's skills at awareness, working, practitioner or expert.
They remember what they've done across projects, learn from retros, and move up a level only when your head of profession reviews the evidence. Run several in the same profession, with different specialisms, and they'll catch different mistakes, as people do.
Senior user researcher
AI practitioner, user-centred design. Specialism: assisted digital.
- User research methods
- Practitioner
- Inclusive research
- Practitioner, new
- Analysis and synthesis
- Practitioner
- Agile research practices
- Working
Remembers
"In the energy grant discovery I under-recruited people who don't use the internet. I now plan those sessions first."
What Civicwright will never do
Knowing where the line is matters as much as what's possible.
Meet your users for you
Practitioners plan research and synthesise it. Your researchers sit with real people.
Pass off rehearsals as evidence
Practice runs with simulated users are labelled and kept out of assessment evidence.
Stand in for a service owner
A named person is accountable for the service, and for every decision in the log.
Sit your service assessment
It rehearses one and assembles the evidence. People assess.
Make policy
It can model options. Deciding what government does stays with officials and ministers.
Questions
Does it replace civil servants?
No. It fills the gaps in a team so the people you have can focus on research with real users, decisions, policy and accountability. Every decision that matters is made by a named person.
Where does our data go?
Civicwright runs on infrastructure you control. Sensitive work is routed only to local models on that hardware. Work sent to Claude goes through the Anthropic API under your own account and settings.
Which AI models does it use?
Claude models through the Anthropic API for judgement-heavy work, and open models through Ollama for routine or sensitive work. Civicwright only offers local models it has tested on your hardware, and picks the cheapest model that meets the quality bar for each task.
Is it endorsed by GDS?
No. Civicwright is independent. It uses the Capability Framework, Service Manual and Service Standard because they're the government's own definitions of good digital delivery.
Can it build services that meet the Service Standard?
That's what it's designed for. Evidence is tagged to the 14 points as work happens, and a mock assessment panel runs at every phase. Passing a real assessment is the goal of the public beta.
What will it cost?
Pricing isn't set. Alpha partners help shape it. Model costs are visible per task, so you'll always see what a sprint costs before it runs.
Private alpha, by invitation only
Civicwright is working with a small number of public sector teams while the platform is built. Joining is by invitation for now.