Frequently Asked Questions

Everything we have been asked enough times to write down. If your question is not here, open an issue and we'll add it.

Last updated: 2026-04-25.

Getting started

How do I install the app on my phone?

Rastrum is a Progressive Web App (PWA), not a native app — there's nothing to install from an app store.

• Android (Chrome / Edge / Brave): open rastrum.org; an "Install" button appears in the bottom-right corner. Tap it. The icon lands on your home screen and behaves like any other app.
• iOS (Safari): Apple does not show an automatic install prompt. Open rastrum.org, tap the Share icon, then Add to Home Screen. The icon shows up alongside other apps.
• Desktop: the same install button shows in Chrome/Edge/Brave on the right of the address bar.

A native iOS App Store wrapper is on the v1.2 roadmap (it requires an Apple Developer account); the PWA covers most use cases for now.

Do I need an account?

No. Rastrum has a guest mode: open the site, hit the camera button, log an observation. It stays on your device until you sign in.

Signing in (magic link, Google, GitHub, passkey, or 6-digit OTP code) gives you:

• Cloud sync — observations live in your account and survive a phone reset.
• Sharing — others can see and comment on your observations.
• Activity feed and badges.
• API tokens for programmatic access.

A "convert guest to account" flow runs the first time you sign in: any local observations get attached to your new account.

Is it free?

Yes. There are no subscription fees, no ads, no upsells. The data you contribute is published under a per-record Creative Commons license you choose (BY 4.0 by default).

Identification

Why don't I see identification suggestions?

The most common reasons:

• No PlantNet key configured. Sign in, go to Profile → Edit, and add a free PlantNet key (my.plantnet.org/account). Without a key, the cascade skips PlantNet and falls through to slower / opt-in plugins.
• Offline-only cascade. If you're offline and don't have any on-device models downloaded, the cascade has nothing to run. You can still log the observation without an ID; the cascade re-runs when you sync.
• Photo too dark / blurry / cropped. Identifiers like PlantNet fail silently on photos with no recognisable subject. Retake closer or with better light.

How accurate is identification?

It depends on the taxon and the photo:

• Plants — PlantNet's top-1 in the cascade is correct ~70-90% of the time on a clear photo of a flower or leaf, less for habit shots or fruits out of season.
• Birds — BirdNET-Lite (audio) is strong on the ~6,000 species in its global model when the call is clean and the species is in range. Photos of birds are harder; we route through Phi-3.5-vision or Claude (BYO key).
• Insects, fungi, lichens, marine life — significantly less reliable. Treat any AI ID as a *suggestion*, not a determination.

The cascade reports a confidence score with every ID. Below 0.4 we flag it as "low confidence" and don't promote the observation to research-grade. Above 0.4 it's still your job to look at the photo and decide whether the guess matches what you saw.

Can I correct an identification?

Yes. On the observation form (or after the fact, on the observation detail page), edit the scientific name field and save. Your manual entry overrides the AI's guess; the original AI suggestion is kept in the identification history for transparency.

If you're a credentialed researcher, your edits to other people's observations carry a 3× weight in the consensus algorithm — see the docs section on expert validation for how to apply.

Audio

Why does BirdNET ask me to download something?

BirdNET-Lite is the on-device audio identifier (Cornell Lab of Ornithology; CC BY-NC-SA 4.0). The model file is ~50 MB. We download it once, the first time you record audio, and cache it locally. After that:

• Identification runs entirely on your phone.
• No audio leaves your device.
• It works offline.

You can clear the cached model from Profile → Edit → BirdNET-Lite → Clear cache if you ever need to re-download.

Why doesn't my recording identify?

Likely causes:

• Model not downloaded yet — the first recording downloads the model; subsequent recordings use the cached copy.
• Species isn't in the global 6,000-species model — BirdNET-Lite prioritises the most-recorded species worldwide. Endemics and regional rarities can be absent.
• Recording too short or too noisy — the model needs at least ~2-3 seconds of relatively clean signal. Heavy wind, traffic, or overlapping voices destroy the spectrogram.
• Wrong taxon — BirdNET only handles birds. Insect, mammal, and frog calls are out of scope today.

Privacy and data

Where do my photos go?

It depends on whether you're signed in and whether you've triggered an identification:

• Signed in: photos upload to Cloudflare R2 (media.rastrum.org) and are linked to your observation row in Supabase. Default visibility is public; you can mark an observation as obscured for sensitive species.
• Guest / offline: photos stay in your phone's IndexedDB outbox until you're online and signed in.
• Identification: photos are sent to PlantNet (when a key is set) and to Anthropic (only when you supply your own key). They are never sent to Anthropic by default.
• On-device models (BirdNET, EfficientNet-Lite0, Phi-3.5-vision, Llama-3.2-1B) never see the network for the data — only the one-time download of the model itself.

See the privacy page for the full breakdown.

Can I delete my account?

Yes. Sign in → /profile/edit/ → Delete account. After confirmation:

• Your auth identity is removed.
• Your observations are anonymised (the link to your user id is severed).
• Photos in R2 are queued for purge within 30 days.
• Audio is deleted immediately.
• Identifications and comments are retained but anonymised.

If you want a full export first, use /profile/export/. If you need observations removed entirely (rather than anonymised), open an issue and we'll work it out.

What if I want my observation to be private?

Two paths:

• Sensitive species are obscured automatically. If your observation matches NOM-059-SEMARNAT-2010 or CITES appendices, the public-readable coordinates are coarsened to ~10×10 km. Only you and credentialed researchers see the precise location.
• Voluntary obscuration. On the observation form, set the "Obscure coordinates" toggle. The same coarsening applies.

We don't have a "private observation" mode (where the observation is invisible to everyone but you) at v1.0. If you have a use case for it, file an issue.

Offline and sync

What happens if I observe without signal?

• The observation, photos, and audio are saved to your phone's IndexedDB outbox immediately.
• A small "pending sync" indicator appears in the avatar dropdown.
• When the next network request succeeds, the outbox flushes: media goes to R2, the observation row goes to Supabase, identifications run on the server.
• If sync fails, we retry on the next online event. You don't have to do anything.

You can keep observing while sync runs in the background.

What happens if I lose my phone?

• Synced observations are safe in your Rastrum account. Sign in on a new device and they show up immediately.
• Unsynced guest observations are lost — they only ever lived in the IndexedDB on the lost phone. This is the one good reason to sign in.

If you sign in *before* losing the phone, anything that synced is durable. If you sign in *after* losing it, the unsynced records are gone.

Technical

Where can I report bugs?

Two places:

• The Report issue button at the bottom-right of every page (it pre-populates a GitHub issue with your locale, page, and basic device info).
• Directly at github.com/ArtemioPadilla/rastrum/issues.

For a more discussion-shaped question (feature requests, "how do I…"), prefer Discussions.

Can I use the API?

Yes. Rastrum exposes a REST API at /functions/v1/api/... and a Model Context Protocol server at /functions/v1/mcp for AI agents.

• Create a personal token at /profile/tokens (token format: rst_*).
• Tokens are scoped (observe, identify, export).
• See the docs index for Module 14 (REST) and Module 15 (MCP). Worked examples for Claude Desktop, Cursor, VS Code, and GitHub Copilot Coding Agent configurations are in those module specs.

Why am I seeing the old version after a deploy?

Usually a stale service worker. Hard-refresh your tab ( Ctrl+Shift+R on Win/Linux, Cmd+Shift+R on Mac) or close and reopen the tab. If you're still seeing the old version, see the service worker cache runbook for diagnosis steps.

Anything else?

Open an issue at github.com/ArtemioPadilla/rastrum/issues and we will reply (and add the answer here if it's a recurring question).

Last updated: 2026-04-25.