● Start · Read before you launch
Costs not included in the price.
What you pay others for, and who pays.
The item price covers the source code, the operator console and this documentation. Running the app needs outside services that bill you directly on your own accounts. This page names each one, says who pays and what drives the bill. It quotes no prices: check each provider's pricing page.
01In one paragraph
For every service below you open your own account, enter your own key in the operator console and pay that provider for your own usage, under its own terms, above any free tier it offers. No API key, account, credit or free quota ships with the download. The code calls no service of the author and pays nobody a fee. The biggest running costs are the AI providers: every chat message, image, edit, video second, voice character and transcription minute is a paid API call on your account. Set spending limits in each provider dashboard and a budget alert in Google Cloud Billing before you launch.
02Service by service
| Service | Needed? | Who pays | What drives the cost |
|---|---|---|---|
| AI chat provider: Anthropic, OpenAI, Google Gemini, or an OpenAI-compatible service (OpenRouter, Groq, DeepSeek, Together, Mistral…) | Required | You, on your own provider account | Billed per token, input and output, at the provider's price for each model. Long chats, the context length you allow (default 30 messages), photo and PDF attachments and reasoning models use many tokens. Web search (off by default) is billed by the provider per search. Chat titles, follow-up suggestions, "Enhance my prompt", Discover briefs and caption translation also use your chat models. A local Ollama endpoint has no API bill but needs your own server. |
| AI image provider: OpenAI, Google Imagen, fal.ai or Stability AI | For the image generator | You, on your own provider account | Billed per generated image at the provider's price; 1 to 4 images per request. |
| Image edit and image tools: OpenAI or fal.ai (edit), fal.ai (remove background, upscale) | Off by default | You, on your own OpenAI / fal.ai account | Billed per edit or per tool call at the provider's price for the model id you enter (Image tools). |
| AI video: fal.ai (Kling models by default) or Replicate | Off by default | You, on your own fal.ai / Replicate account | Billed per second of video or per clip, at the provider's price for the model. Video is usually the most expensive module per request. The default cap is 50 jobs a day for all users together (Video Studio). |
| Natural voices (text to speech): OpenAI, ElevenLabs or Gemini | Off by default | You, on your own provider account | Billed per character read aloud. Voice mode reads every answer, so it uses characters quickly. Without a voice the app uses the free device voice (Voice). |
| Transcription: OpenAI, Deepgram or Gemini | Off by default | You, on your own provider account | Billed per minute of audio for voice notes, the voice studio and auto captions. Dictation and voice mode listen with the phone's own speech recognition, which costs you nothing. |
Moderation endpoint (OpenAI omni-moderation-latest) | Optional, off by default | You, on your own OpenAI account | Uses your OpenAI key. The console labels it free, as OpenAI priced it when this version was written; check OpenAI's pricing page. |
| Discover feeds: RSS / Atom feeds of publishers you choose | Off by default | No fee to fetch; the briefs use your chat provider | Fetching a public feed is free, but each publisher's terms decide whether you may show its headlines in your app, and some require a licence or forbid commercial use. Every default feed ships off until you have read its terms (Discover → publisher terms). AI briefs are billed by your chat provider. |
| Firebase / Google Cloud, Blaze plan: Cloud Functions, Cloud Scheduler, Firestore, Storage, Authentication, Cloud Messaging | Required | You, on your own Google Cloud billing account | Blaze is pay-as-you-go above Google's free quota. Here: function run time (a chat stream keeps the function busy until the answer ends, up to 300 s), Firestore reads and writes (usage counters, statistics, settings, Discover items), Storage for generated images and videos and their downloads, three scheduled jobs, outbound network traffic to the providers. Phone sign-in by SMS is not used. |
| Console hosting and a domain | Required (domain optional) | You, at your own host and registrar | One small always-on Node 22 server with a persistent disk (VPS, Render, Railway, Fly, a container, Cloud Run), or a serverless host plus a Postgres database (SQLite or Postgres). HTTPS is required in production. |
| Apple Developer Program | To publish on iOS | You, yearly membership | Needed to sign and publish the iOS app, sell in-app purchases, send push (APNs key) and offer Sign in with Apple. |
| Google Play Console | To publish on Android | You, one-time registration | Needed to publish the Android app and sell in-app products. |
| Store commission on purchases | Only if you sell plans, credits or Remove ads | Deducted by Apple and Google from each sale | Apple and Google keep a commission on in-app purchases and subscriptions, at their current rates for your programme. The console's Revenue centre records the price the store reports; it does not deduct the commission. |
| Ad networks: AdMob, Unity Ads, ironSource LevelPlay | Only if you show ads | Your own publisher account; nobody pays to join, the network keeps a revenue share | The networks pay you for ads shown and keep their share under their own terms. They approve your app and account first; payouts follow their thresholds and schedules. Ads ship switched off. |
| Optional: Postgres for the console | Serverless or several instances | You, at your own database host | Only when you set DATABASE_URL. One server with the built-in SQLite file needs no database. |
| Optional: Firebase Hosting for your own API domain | Optional | You, on Blaze | Only if you serve the API at https://your-domain/api with a Hosting rewrite. |
| Optional: Vercel for a public web demo | Optional | You, on your own Vercel account | Only if you publish a public web demo (Deployment → hosting a public demo). |
03Keep the bill predictable
The backend has limits built in. The defaults are low on purpose; raise them when you know your costs.
| Control | Default | Where |
|---|---|---|
| Free allowance per user | Per day: 20 messages, 2 images, 2 image edits and tools, 3,000 natural-voice characters, 5 transcription minutes, 5 Discover briefs, 3 rewarded unlocks. Per month: 0 AI video seconds, 3 auto-caption minutes | Console → Limits |
| Allowance of each plan | Default plan "Pro": 500 messages, 50 images, 50 edits, 100,000 voice characters, 120 transcription minutes, 100 briefs a day; 60 video seconds and 60 caption minutes a month. Plans are never unlimited. | Console → Plans |
| AI calls per day, all users together | 2000; above it the API answers 503 capacity until 00:00 UTC | Console → AI providers & keys |
| Images per day, all users together (edits and tools count too) | 200 | Console → Image providers & keys |
| Video jobs per day, all users together | 50 (at most 2 unfinished jobs per user) | Console → Video (settings) |
| Natural-voice characters / transcription minutes per day, all users together | 1,000,000 / 1000 | Console → Voices & transcription (settings) |
| Per-user rate limit | 30 chat and 6 image requests per minute | Fixed in the backend |
| Credit cost per model | Chat: Muse Fast 1, Muse Pro 3, Muse Max 8. Images 1–2. Edits 2. Tools 1. Video: credits per second × duration (defaults 10 and 6 per second) | Console → Chat models, Image models, Image tools, Video |
| Max output tokens per answer | 4096 (hard maximum 32000) | Console → Chat models |
| Cloud Functions instances | maxInstances: 20 | firebase/functions/src/index.ts (code change) |
- Set a spending limit or prepaid balance in each provider's dashboard (Provider dashboards).
- Create a budget alert in Google Cloud console → Billing → Budgets & alerts.
- Enter your own prices in the console (per model, per voice, per video second, per transcription minute, per edit and per tool call). The Usage & spend page then multiplies the measured usage by your prices. It is labelled "computed from usage" and is not your provider's invoice.
04Testing for free
You can try everything without paying anyone: the local API in demo mode, the web app and the console with sample data (Quick start), and the Firebase emulators (Installation). Real AI answers start costing money on your own account as soon as you add a provider key.