Documentation

Guides

Quick start, features, shortcuts, credits, models, privacy and fixes — written in plain language.

Quick start

From signing in to a finished result in five minutes.

1. Describe the goal, not the tool: Open a new Super and type what you want in plain words — “write a launch email for my skincare brand”. AIR picks the model, the research depth and the tools for you.
2. Attach anything relevant: Drop PDFs, spreadsheets, images or audio into the composer. AIR reads them and uses them as context instead of guessing.
3. Pick a mode when it matters: Modes (compass icon) hold Mission mode for multi-step work AIR runs end to end, and Advisor mode for an honest reality check on a plan.
4. Turn a thread into a project: Projects keep a goal, tasks, memory and a timeline in one place, so long work survives across days.
5. Review, improve, export: Use the tools under each reply to improve it, get a second opinion from a different model, or export it as a document.

Feature guide

What each surface is for, in one line each.

Supercomputer: The main conversation. One place to ask, create, research and finish work.
Projects: Goal-driven workspaces with tasks, memory, timeline and next actions.
Mission mode: Give one outcome; AIR plans, runs the steps and reports back.
Advisor mode: Adversarial review: risks, weak assumptions, feasibility and cheaper paths.
Creations & Library: Every image, video and voiceover AIR made, searchable and re-downloadable.
Action center (Jobs): Running and finished background work, with failures and retries.
Workflows & Skills: Repeatable chains you can re-run instead of re-prompting.
Health check & Status: Audit of your workspace, plus live model and provider health.
Memory & Privacy center: See what AIR remembers about you and delete any of it.

Keyboard shortcuts

The handful worth memorising.

⌘K / Ctrl+K: Open the command palette — jump to any page or start any creation.
/: Opens the command palette too, whenever you are not typing in a field.
Enter / Shift+Enter: Send the prompt / add a new line without sending.
Esc: Close the palette, a popover or a dialog.
Hold 3s on a sidebar item: Opens its context menu (rename, pin, add to project, delete).

Credit guide

Five credit types, each spent only on what it powers.

Text credits: Chat, research, writing and analysis. Short prompts cost about 1 credit; long multi-step work costs more.
Image credits: Image generation and editing — roughly 9 credits per standard image, more for high resolution.
Voice credits: Voiceovers and text-to-speech — roughly 9 credits per standard run.
Video credits: Avatar and faceless video. Pricing scales with length and quality.
Cloud storage: Capacity, not per-run spend. Your generations and uploads live here.
If something fails: Failed generations are refunded automatically. If a charge looks wrong, the Help center has a billing contact.
Full pricing detail: The Documentation page breaks down every plan, top-up and per-credit cost.

AI model guide

You should not have to choose — but here is how AIR chooses.

Automatic routing: AIR reads the request and routes it: fast models for short answers, reasoning models for analysis and code, research pipelines for cited questions.
Why this model?: Hover the engine chip in the composer to see the plain-language reason behind the current routing.
Air Max: The orchestration engine: it plans, splits work across specialists, checks the result and reports back.
Second opinion: Any reply can be independently audited by a different model family to catch errors and overclaims.
Overriding: You can still pick an engine manually in the model selector when you want a specific behaviour.

Privacy guide

Your work stays yours.

What is stored: Your conversations, projects, files and generated assets — so you can come back to them.
Memory: Memory is scoped and optional. The Privacy center shows every stored fact and lets you delete any of them.
Turning memory off: Memory can be disabled entirely; AIR then relies only on the current conversation.
Deleting data: Delete a conversation, a project, a file or all memory from the Privacy center at any time.
Connectors: Connected accounts only use the permissions you granted, and can be disconnected from the Connectors page.

AI agents

Your AI company: hire a worker, design how it works, let it run.

1. Hire your first agent: Open Agents and tell your CEO what you need done. They hire the right worker, give it a name, a face and a job — you don't configure anything to begin with.
2. Open its canvas: The Canvas link on an agent's profile shows how it works as a chain of steps: what starts it, what it thinks about, what it makes, and what it does with the result.
3. Start from a template: Content creator, Research desk, Inbox assistant and Social engine all arrive with instructions already written. Change any step's wording to make it yours.
4. Build by asking: The box at the bottom of the canvas edits the workflow for you — “add a researcher before the writer”, or “this isn't working, fix it”. You see exactly what changed and choose whether to apply it.
5. Decide what it may do alone: Drafts only keeps everything inside AIR. Ask me first lets it work freely but stops before publishing, sending or spending. Work on its own removes the gates, inside your caps.
6. Schedule it: Give the trigger a shift — every two days at 4pm, every Monday morning. Agents run in the background, so closing the app changes nothing.
How it gets better: After each run the agent scores its own work out of 100, writes one lesson, and rewrites its weakest step. Those lessons are read back at the start of every following run.
The supervisor: The layer above your team. It reads what everyone actually did, gives each agent one honest verdict — on track, needs you, or stalled — and with repair on, rewrites a stalled agent's workflow and puts it back to work.
Approvals: When a run stops for you, it's waiting in the Runs panel on the canvas. Approve the step and the rest of the run carries on from there.
What it costs: The canvas header shows the credits per run before you activate anything, and a monthly cap stops the agent cleanly once it's reached.
Connected accounts: A step that needs YouTube, Gmail, Drive, Slack or Notion says so on the canvas instead of failing silently — connect the account from Connectors.

Troubleshooting

The issues people actually hit, and the fix.

A reply came back empty: Usually a provider hiccup. Re-send; AIR retries on a different model. Persistent cases show on the Status page.
Research returned nothing useful: Add a time window (“in 2026”) or a source type (“from official docs”). Explicitly saying “search the web” forces fresh retrieval.
A job seems stuck: Open the Action center. Background work continues even if you close the tab, and failures list a reason.
Credits look wrong: Check Usage & insights for a per-category breakdown. Failed runs refund automatically.
Something looks broken visually: Reload once; if it persists, send a bug report from the Help center with the page URL.
Lost work: Unsent prompts and drafts are restored when you return to a Super — check the sidebar for the recent session.

Still stuck? The Help center has FAQs and a direct line for bugs and billing, and Documentation covers plans and pricing in full.