Say It Once. Your AI Remembers Forever.

Stop starting every session with “use British English, follow our conventions, don’t touch the migrations”. Write your standards once as a ruleset, attach it to your projects, and every AI assistant follows them automatically.

The same instructions, every single session

Without rulesets
  • You retype the same standards at the start of every session
  • Each teammate’s AI works to slightly different conventions
  • A standard changes and nobody updates the six places it lives
  • You find out a convention was missed in code review, not before
  • Nobody can say which instructions an agent was actually given
With rulesets
  • Written once, followed on every project you attach them to
  • The whole team’s assistants work to the same standards
  • Change a rule in one place and it updates everywhere
  • Hard prohibitions that a project cannot quietly opt out of
  • A view of exactly which rules are in effect, and why

Four types of rule

The type carries the weight. It is the difference between an instruction your agent will stop and ask about, and one it will sensibly work around.

Must

When you’d say "always", "make sure you"

The agent complies, or stops and asks. Use it when there is no acceptable alternative.

Must use our i18n functions for all user-facing strings

Must not

When you’d say "never", "under no circumstances"

A prohibition. In the workspace baseline, this is the one type nothing can override.

Must not commit credentials or environment files

Prefer

When you’d say "I’d rather", "unless there’s a reason"

A preference with an honest escape hatch. Reach for this more than you expect.

Prefer date-fns over Moment.js

Consult

When you’d say "check our X before doing Y"

Points at a document, loaded only when the work calls for it. Cheap to carry.

Consult the design system before building any new UI

One baseline for the things that are never negotiable

Every workspace has exactly one baseline ruleset. It reaches every project automatically, cannot be detached, and is capped at twenty rules.

BaselineOrdinary ruleset
How manyExactly one per workspaceAs many as you like
Applies toEvery project, automaticallyOnly where you attach it
DetachableNoYes, instantly
Rule cap2050
What belongsSafety and complianceConventions and house style

A project can override a baseline must or prefer — a docs project might legitimately swap British English for US spelling. A baseline must not can never be overridden by anything. A team can reasonably say “we do it differently here” about a convention. It should not be able to say it about “never log customer data”.

Built to be trusted, not just followed

See what is actually in effect

Every rule reaching your agents, grouped by ruleset — plus what is not applying and why: duplicates, rules replaced by a more specific one, archived rulesets, permissions. The fastest answer to "why did my agent do that?".

Versioned and attributed

Every change is saved as a version and recorded in your activity feed, including which rules an AI wrote. You can always see what changed, who changed it, and undo it.

Your AI can author them

Say "remember we always use British spelling" and the rule gets written, typed and reasoned for you. It cannot touch your baseline, and cannot delete or rewrite rules it is working under.

Scope to where it matters

Narrow a rule to a path like apps/admin/** so it applies only where it should — and so a human can check whether it was honoured.

Point at documents, do not paste them

A consult rule links to a context library instead of inlining it. The pointer is always present and costs almost nothing; the document loads only when the work calls for it.

Honest about enforcement

A rule is only marked Checked when you name a real lint rule, CI check, test or workflow gate. Everything else is Advisory — because claiming a rule is enforced when nothing enforces it is worse than saying nothing.

Common questions

What is a ruleset?

A named group of standing instructions your AI follows. Write them once, attach the ruleset to any project, and every assistant working there follows them — without you repeating yourself at the start of each session.

What is the workspace baseline?

One ruleset per workspace that applies to every project automatically, cannot be detached, and is capped at 20 rules. It is where your non-negotiables live. If forgetting it once could cause an incident, it belongs here.

Can AI write rules for me?

Yes — tell your assistant to remember something and it will create the rule, pick the type and record the reason. Every AI-authored rule is logged and versioned. It cannot touch your baseline, and cannot delete or rewrite existing rules.

Will this eat my context window?

Rules are short by design, which is why the limits exist. For anything longer than a sentence, a consult rule points at a document that loads only when relevant — the pointer costs almost nothing.

How is this different from a CLAUDE.md file?

A CLAUDE.md lives in one repository. Rulesets live in your workspace, attach to many projects, and change in one place — plus they carry rule types, path scoping, version history and a record of who changed what.

Write it down once

Every workspace starts with a baseline and an example ruleset, so there is something to edit rather than a blank page.

Start free