Manifesto

The shared principles, in the words of draft 4. Everyday Open is the name of the umbrella. The title underneath is the title of this draft.

Section headings are in sentence case. Principle lines are word for word. The draft still says Essex Garden is being planned. It is now in preview, and that sentence is left as written. The plain text is MANIFESTO.txt.

OPEN KNOWLEDGE, PUT TO WORK

An umbrella manifesto (draft 4, for the community to tear up and rewrite)

Why

Human knowledge is mostly free to read but hard to use. AI can help close that gap by turning trusted knowledge into practical help that helps people do more. We want that help to be free of cost, adverts and surveillance, and owned by the people who use it.

Using AI has a real water, energy and carbon cost. We treat that cost as a design constraint, not a rounding error: we measure it, we cut it first, and we only then fund credible restoration or efficiency work for what is left. We do not pretend a credit makes the use disappear.

What this is

An umbrella for community-led projects that share one set of principles. Each project solves a concrete problem. Together they share standards, tools and lessons, so that every new project starts further ahead than the last.

Projects under the umbrella

Projects are added as the community proposes and builds them. A project joins when it adopts these principles and has a working first version. The umbrella must not grow faster than its projects actually work.

  • First candidate (not yet a member): How to look after an Essex garden, a books-to-tasks website, being planned. It will turn trusted public-domain books into sourced, personalised tasks, with a full-text reader and on-device data by default, and it will link back here.
  • Future examples: offline tutors, practical skills guides, home and energy planning, repair guides, and civic and political understanding.

Principles (shared by every project)

1. Open source, always.

Code, prompts, rules and data formats are published under open licences. Anyone can inspect, fork or self-host.

2. No adverts. No selling data.

User data stays on the user's own device. Nothing leaves it unless the user chooses to share, and they can see exactly what would go.

3. Free to use, honest about cost.

AI isn't free to run, so we're open about who pays. By default it runs on small open models on the user's own device. Optional extras are clearly labelled: bring your own API key, or use donation-funded hosting that is charged at cost and has published accounts.

4. Point to what already works.

If Wikipedia, Khan Academy, Kiwix, official guidance or another open project already does it well, we link to it or build on it rather than rebuild it.

5. Every claim shows its source.

Outputs link back to where they came from. If we can't cite it, we don't present it as fact.

6. Safety before convenience.

Anything that could cause harm carries a warning. Outdated advice is flagged as "Old advice" with a link to current guidance. Advice we can't verify is marked "Unverified", and the "Just do" mode is switched off for it.

7. Learn at your own depth.

Wherever there's something to learn, users choose how they engage:

  • Just do: clear steps, ready to follow.
  • Do while learning: the steps, plus the why.
  • Just learn: the knowledge itself, without the task.

8. Accessible by default, consistent everywhere.

Every project uses the same accessibility settings and meets WCAG 2.2 AA as a minimum. The community suggests improvements and tests them with real users.

9. Respect creators.

We only use material we have the right to use. Sources are credited and linked.

10. Every change is traceable.

Every edit records who made it, when, why, its source and exactly what changed, and it can be undone. Safety-critical edits need two reviewers. Disagreements are recorded openly, not hidden.

11. Governed in the open.

Decisions follow a written process that anyone can read and challenge. Founders don't get the final say forever; authority passes to the community as it grows.

12. Reduce first. Offset only what remains, and never greenwash.

We estimate and publish the water, energy and carbon used to build and run each project, with the method and the uncertainty. We avoid unnecessary model use, prefer small, efficient and local models and short context, and only then fund credible, place-specific restoration or efficiency work. Offsetting water is weaker than using less, and a litre "replaced" in one catchment does not cancel a litre consumed in another. We never claim to be water-neutral or carbon-neutral on the back of weak credits.

What the umbrella provides

  • These shared principles, and a simple way to check a project meets them.
  • Shared building blocks: accessibility settings, the three learning modes, safety warnings, source citation and traceable edits.
  • A shared way of working, including the multi-bot development framework (vibecodebots), so that small teams can build to a high standard.
  • A shared method for estimating, reducing and reporting the water, energy and carbon of AI use, and for choosing any later offsets (SUSTAINABILITY.txt).
  • A place to propose projects, find contributors and learn from other projects' mistakes.

How we'll know it's working

  • People use what we make to actually do things they couldn't do before.
  • No unsafe instruction ever reaches someone in "Just do" mode.
  • Everything runs on cheap, older devices, including offline.
  • New projects reuse shared parts instead of starting from scratch.
  • People choose to contribute without being paid.
  • Each project publishes its estimated water, energy and carbon, with method and uncertainty; those estimates fall over time per unit of useful work; offsets are listed separately and are never used to net a headline to zero.

What we won't do

  • Grow by advertising, tracking or engagement tricks.
  • Present AI output as settled truth.
  • Let the umbrella grow faster than its projects actually work.
  • Claim water or carbon neutrality, or "offset AI", on the basis of credits that are not additional, verified and like-for-like.
  • Let buying offsets replace cutting use.

This is a first pass. It will change. That's the point.