28 Intelligence · In development
Intelligence that's woven in,
not bolted on.
One shared AI layer that every product uses: to understand what people are trying to do, suggest the next useful step, and carry out approved work. Powerful, but always visible and always in your control.
Model-independent
Products talk to 28 Intelligence, not to one model. As better models arrive, we switch without rebuilding apps.
The right model for the job
Quick tasks get fast, efficient models; hard ones get the most capable. Chosen by complexity, speed, cost and reliability.
Agents with permission
Specialized agents research, design, code, test and operate, using only the tools and data they've been granted.
Memory you can see
It remembers approved project history and preferences. You can inspect, correct or clear what it knows at any time.
Evidence, not guesses
Every suggestion shows why. When the signal is thin, it says so ("Early signal, based on 40 visits").
Ask before it matters
Public releases, spending, messages to other people and anything destructive always need a person's approval, after a preview.
Our guardrails
Autonomy that stays observable and reversible.
No silent changes
AI never modifies production code or live data on its own. Changes go through preview, tests and approval.
Least access
Agents see only what a task needs, never unrelated private information.
A full record
Every decision, tool call and approval is logged and can be reviewed.
Works when AI doesn't
Editing, previews and publishing keep working if a model is slow or unavailable.