FutureInvokedIndependent software studio
← All servicesService / 02

AI integration with human controls

AI tools
with a person in control.

Research and drafting tools that connect models to useful context and keep the important decisions reviewable.

Email us the task ↗What to put in the email ↓
15 s · Sample and demo data
AI can draft. It should not decide.The pattern we build: a check and an owner for every AI step.Title 'AI can draft. It should not decide.', a four-step diagram (the model drafts, code checks, a person approves, a trail remains), a research tool where draft angles keep their source tags, and a card of rule results produced by code. Ends on the futureinvoked.uk card.

THE PROBLEM

AI is useful for research and drafts. It should not make the decisions.

Language models are good at gathering material, summarising and drafting. They are also sometimes confidently wrong. We build tools that put a model to work on your own context, keep the sources attached to what it produces, and stop at a review step before anything is sent, saved or acted on. Where data should not leave your computer, a local model can do the work.

  • Research assistants and source-backed drafts
  • Model connections and local inference
  • Approval workflows and review controls

Tasks that fit

  • Drafts written from your own documents, with the sources shown
  • A regular brief assembled from several systems
  • Incoming requests sorted and answered in draft for a person to approve
  • Agent work that needs limits and a record of what happened

What you get

The model does the legwork. A person decides.

  • A model connected to the context it needs
  • A review step before results are used
  • Where it matters, a record of what was produced and approved
  • A quote agreed before any work begins

FROM OUR WORK

See similar work.

Short videos and project pages from work we have already built.

12 s · Sample data
Research first. Then write.Signal Drafts: findings that keep their sources.Title 'Research first. Then write.', then a research tool's evidence review moving to its key findings, each with a source tag, with the first finding highlighted. Ends on the futureinvoked.uk card. Sample data.Signal Drafts →
13 s
Make the work visible. Verify the result.Control Room: the six parts of the system.Text animation for the Control Room project: its tagline, then a numbered list of the six parts of the system, from task request to operator view, then the FutureInvoked end card with a one-line summary.Control Room →
  • Signal Drafts carries source evidence from a research question into drafts, and records the exact text that was approved.
  • Control Room gives agent work an explicit task lifecycle, a separate verifier and an inspectable trail.
  • Hermes Operator Controls adds tool budgets and typed verification records to an AI agent.
  • Forensic Lens runs a model locally inside a browser extension, on the pixels it captures.

GUIDES

Read first.

WORKING TOGETHER

Start with the problem.

Four steps. Nothing is built before the scope and the quote are agreed.

  1. 01

    Share your brief

    Email the problem, who the tool is for and what you need it to do. Existing tools or examples help.

  2. 02

    Define the scope

    Discuss requirements, constraints and deliverables. Agree a quote and approach before work begins.

  3. 03

    Build & review

    Review the work against the agreed scope and give feedback as the application takes shape.

  4. 04

    Delivery & handover

    Receive the agreed deliverables. Discuss setup, documentation and any further support your project needs.

QUESTIONS

Before you write.

Will the AI make decisions on its own?
No. The tools we build gather, sort and draft. A person reviews before anything important is sent, saved or acted on.
Can our data stay private?
It can. Where data should not leave your computer, a local model can do the work. Otherwise we agree which provider sees what before anything is built.
Which AI models do you work with?
Hosted models such as Claude and OpenAI's, and local models that run on your own machine. The choice depends on the task, the data and the running cost.
What does an AI tool cost?
Each tool is quoted after we read your email and agree the scope together. If a hosted model is involved, its running cost is part of that conversation.
How do we start?
Send one email with four things: the task, the tools you use today, what a good result looks like and when you need it. We answer by email, with questions or with a scope and a quote.

START WITH ONE EMAIL

Tell us about
the task.

Four lines are enough. We answer by email, with questions or with a scope and a quote.
There is no form to fill in and nothing to sign up for.

Email the studiofutureinvoked@proton.me

WHAT TO PUT IN IT

  1. The taskWhat is done by hand today, and how often.
  2. The toolsWhere the data lives: files, sheets, apps.
  3. The resultWhat a good outcome looks like for you.
  4. The timingWhen you need it.
Write the email

The button opens your email app with the four questions already in the message. Examples: how to write a software brief.

Pricing on requestA quote is agreed before work beginsBased in CzechiaRegistered business: details on the imprint page →