System Overview

Purpose

ManyAI is a React Native mobile application that acts as a multi-provider AI router. It intelligently routes user prompts to the best available free AI provider, implementing an automatic fallback chain that cycles through up to 8 providers before giving up. The core promise: you always get an answer, even when individual providers are rate-limited or down. It is shareware — free to use, supported by donations.

Tech Stack

Component	Technology
Framework	React Native (Expo SDK)
Routing	Expo Router v3 (file-based)
Language	TypeScript
Storage	AsyncStorage
Image handling	expo-image-picker
OTA updates	expo-updates
Version	1.0.5

Navigation (3 Tabs)

The app uses a bottom tab navigator (expo-router Tabs). Tab bar is dark-themed (#16213e background, #4ECDC4 active tint).

Tab	File	Purpose
Home	`app/(tabs)/index.tsx`	Main chat interface
Saved	`app/(tabs)/saved.tsx`	Saved responses management
Settings	`app/(tabs)/explore.tsx`	API keys, providers, models

Provider System (13 Providers)

Providers are defined in lib/providers/providers.ts. The routing order (ROUTING_ORDER) defines priority. pickProvider() first tries task-best providers, then any capable provider, then falls back to Pollinations.

Provider	Free Tier	Vision	Key Required	Notes
Cerebras	Yes	No	Yes	Fastest inference
Groq	Yes	No	Yes	Fast, reliable general
Gemini	Yes	Yes	Yes	Best for summarization, translation
Mistral	Yes	No	Yes	Best for coding, creative
SambaNova	Yes	No	Yes	Best for deep reasoning
OpenRouter	Yes	No	Yes	Multi-model gateway
Hugging Face	Yes	No	Yes	Massive model selection
Cohere	Yes	No	Yes	Strong summarization
Cloudflare AI	Yes	No	Yes	Key format: `accountId:apiToken`
Fireworks	Yes	No	Yes	Strong with DeepSeek V3
OpenAI	Yes*	Yes	Yes	*Requires billing setup
Anthropic	No	Yes	Yes	Paid only, no free tier
Pollinations	Always	No	No	Ultimate fallback, always available

Vision-capable providers (accept image input): Gemini, OpenAI, Anthropic

Key Features

Auto-fallback: Up to MAX_RETRIES = 8 provider attempts per message
Session failure tracking: Failed providers are skipped for the rest of the session (failedProviders useRef)
Conversation history: Last 10 messages sent as context (excludes system/error messages)
Image input: Attach from gallery (🖼) or camera (📷); base64 encoded, quality 0.5; vision providers only
Image generation: Keyword detection (isImageGenRequest()) routes to image providers (Pollinations, free)
Save responses: Any AI reply can be saved to a user-defined category
Refine flow: Saved responses can be sent back to Chat to continue the conversation
OTA updates: Silent background check on every launch via expo-updates; prompts to restart if update available
First-launch onboarding: Help modal shown automatically on first open, dismissable, accessible via /help or ? button
Remote config: stevepleasants.com/manyai/config.json can patch model lists without a new build

Storage (AsyncStorage Keys)

Key	Contents
`manyai_keys_*`	API keys per provider
`manyai_provider_order`	User-customized provider priority
`manyai_enabled_*`	Per-provider enabled toggle
`manyai_model_*`	Per-provider selected model
`manyai_saved_*`	Saved responses by category
`manyai_categories`	List of user-created categories
`manyai_onboarded`	Flag — hide help modal after first launch

Routing Logic (`pickProvider`)

1. Build available = keys present + pollinations (always)
2. If image attached: filter to VISION_PROVIDERS only
3. First pass: iterate ROUTING_ORDER, skip excluded/disabled,
               prefer bestFor[taskType] match
4. Second pass: iterate ROUTING_ORDER, return first capable provider
5. Final fallback: return 'pollinations' if candidate
6. If all exhausted: return null → show error message

API Shapes Supported (`callProvider.ts`)

Pollinations: GET text.pollinations.ai/{encoded_prompt} — no auth, no key
Gemini: POST /models/{model}:generateContent — parts array with inline_data for images
Anthropic: POST /messages — x-api-key header, different body structure
Cloudflare: Account ID embedded in URL, key split on : separator
All others: OpenAI-compatible /chat/completions with Authorization: Bearer header

Request timeout: 30 seconds per attempt.