Everything a new Loom user needs — install, sign in, and the first few things worth doing.
Download the Mac (Apple Silicon) build from stera.se/loom/download. Double-click the .dmg to mount it, then drag Stera Loom into Applications. Eject the installer and open Loom from Applications.
Loom is signed and notarized by Apple, so macOS opens it without the usual unidentified-developer warning. Windows and Linux builds follow shortly — the cloud backend works the same on every platform; only Stera Mini (the local model) is Apple Silicon for now.
On first launch, click Sign in with Stera. Your browser opens stera.se, you authenticate, and Loom picks up the session automatically. If you don't have an account, create one at stera.se/account.
Loom never stores third-party credentials. It holds your Stera session token to reach our backend; all other logins (Gmail, GitHub, etc.) are delegated to your real browser via Stera's managed Chrome profile — Loom drives the browser, it doesn't store passwords.
Loom is pay-as-you-go. There are no monthly tiers, no "premium voices", no gated models. Every provider bills at 1.4× the real upstream cost, to the microcent. What you see in your usage panel is what you actually spent.
First run with a zero balance will open Settings with an Add balance banner. $5 goes a long way on text chat. Your balance never expires and you can top up or stop any time.
Build apps. Type what you want in the Loom window — a bookmark manager with tags, a freelancer invoice dashboard, a recipe planner. Loom plans, writes, and runs the app locally. Built apps live in Documents/Loom Apps and run on your machine only.
Run Fabric — your personal AI center. Ask Loom to build "Fabric" and you get a conversational AI that lives on your machine: chat, notes, workspaces, cross-workspace awareness. Fabric calls tools (email, web, calendar) through Loom, so your accounts stay in your own browser.
Drive your browser. Loom can launch a Stera- managed Chrome window (dedicated profile, kept separate from your personal one) and operate it on your behalf — reading pages, filling forms, pulling data back. You stay in the driver's seat for anything high-stakes.
Every window shows today's spend in the top-right corner. Click it to see a 14-day breakdown, per-model costs, and a balance summary.
Two controls matter for pay-as-you-go. A monthly cap stops AI calls once you've spent a set amount this month. A pause switch stops all AI spending immediately — useful if you spot something unexpected. Both live in the usage panel; both can be lifted with one click.
Stera Mini is a small Stera-trained model that runs fully on your Mac's GPU — no API calls, no cost. It's good for quick replies ("thanks", "translate this", "rewrite this shorter") where cloud round-trips would feel sluggish. It's not the right tool for coding or long reasoning — for that, keep routing to cloud.
Settings → Stera Mini → Install + Load. The first load downloads ~5 GB from our CDN and compiles Metal shaders once (~2 min). After that, each launch takes a few seconds. Mac Apple Silicon only for now.
Voice input. In Fabric, the microphone button in the input bar captures speech continuously and transcribes it live. Works offline — we use your browser's speech recognition.
Voice output. The speaker icon in Fabric's header turns on spoken replies (system voice, zero cost). You don't need the icon if you're on bluetooth — say "switch to voice" or "stop voice" and Fabric handles it. Paid voices (ElevenLabs, OpenAI) are available per- message via the synthesize_voice tool when you ask for them.
Your conversations, notes, files, and built apps stay on your machine. Stera's backend is a stateless proxy to LLM providers — we bill, route, and return the response. We don't store conversation content. Full details in the privacy policy.
Third-party accounts (Gmail, GitHub, etc.) are held by your browser, not Loom. When Fabric needs to read email or open a page, it drives the browser — the credentials never leave your keychain.
Loom won't open. Right-click the app in Applications → Open. If macOS still blocks it, make sure you're on macOS 13+ with Apple Silicon.
"Insufficient balance" during a chat or build. Top up in Settings → Balance. If you have a monthly cap set and you've hit it, raise or remove the cap in the usage panel.
Stera Mini won't load. First-load shader compile can take ~2 minutes on a fresh machine — wait it out once, then later launches are fast. If it still fails, try Stop & Unload then Start again in Settings.
The AI skipped a tool call. If a reply says "Skipping" or the model refuses a modality task, try once more — Loom's self-repair layer will escalate to a more reliable model automatically. If it persists, pick Kimi or Claude Sonnet in Settings → Model for that session.
Questions, bug reports, feature requests — email support@stera.se. Include a short description of what you were doing and the model you were on; that's usually enough for us to reproduce.
These docs will grow with the product. Check back after updates — new capabilities show up here first.