AI chat is great — until the work gets serious.
Then it drifts — scope creeps, decisions get buried, review gets skipped. AIR — short for AI Resource — gives the session scope, structure, one active step at a time, benchmarked delivery, and resumable handoff.
Same chatbot. Different session.
AIR changes the session, not the model — so the same work stays scoped, reviewed, auditable, and resumable.
Configure, then execute
Map the work before acting.
Honest by design
Surfaces uncertainty, never hides it.
Benchmark before delivery
Held to a task-fitted standard.
Yours anywhere
Portable. No provider lock-in.
A teammate, not an agent.
AIR is not an autonomous agent, and it is not built for hands-off automation. AIR is built to work with you, not for you. A few habits get the most out of it:
Agree up front what's yours to own and what's AIR's — a clear split keeps the work focused.
When AIR is unsure it asks instead of guessing — give it the direction or sources it needs.
On the hard parts, let AIR lead — it catches adjacent risks and blind spots. Steer or overwrite anytime.
Start where you need to.
AIR is a structured generalist. AIR brings discipline to code, research, writing, strategy, analysis, and execution without dependencies.
How it works
The model behind AIR — onboarding, the orbit model, gates, and handoff continuity.
See the model →Get started
Grab the boot bundle, start a session, answer a few questions, and you are configured.
Start a project →Use cases
Development, research, writing, strategy — one framework, many kinds of work.
See the range →Principles
Prompt-compiled, fail-closed, judgment in the loop. Why AIR won't oversell itself.
Read the stance →AIR is prompt-based, so it can't guarantee correctness. Its output depends on the model you run it on and can vary. AIR adds structure and surfaces uncertainty — it doesn't replace your judgment, so check what matters.
AIR costs you something.
It adds structure, and structure isn’t free. Three things to weigh before you reach for it.
AIR scopes, gates, and reviews, so a session runs more tokens and more turns than raw chat. That overhead is the price of the structure.
For a one-off question, raw chat is faster and lighter. AIR is built for sustained, serious work — reach for it when drift and unreviewed output would actually cost you.
The gates, benchmarks, and handoff model take a session or two to feel natural. The payoff lands once the workflow is a habit.
The proof is the work itself.
No fabricated testimonials. AIR’s own brand, site, and recovery were run as structured AIR projects — so the clearest evidence it works is what built and repaired this very site.
Built with AIR
How AIR’s brand, site, and framework materials were produced as a scoped, reviewable AIR project.
Read the proof path →Recovered with AIR
Restoring a project after a session went wrong — scope re-grounded, decisions re-verified, continuity rebuilt.
See the recovery →Real boot vs roleplay
A model can accept the files and perform AIR without running it. Here’s how to tell a real boot from the performance.
See the tell →See a full session
One real session, end to end: booted on Claude Opus 4.8, it frames a risky project, settles the riskiest decision first, and hands off — every object it emitted shown in full.
Watch it work →Your honest take
No outside reviews yet — and we won’t fake them. Used AIR on real work? Tell us what held and what didn’t.
Share your experience →Bring your model. AIR brings the structure.
Grab the boot bundle, start a session, and answer a few questions. You'll have a configured project in minutes.