What is Reality Fabricator?

Reality Fabricator is an AI platform for building autonomous agents, playing interactive fiction adventures, chatting with AI companions, and creating media (images, video, VTuber avatars). It supports 15+ AI models including GPT-5.6, Claude Fable 5, Gemini 3.1, Grok 4.5, and DeepSeek V4, plus free models like Llama 3.3 70B. Available in 6 languages with encrypted local storage.

Is Reality Fabricator free?

Yes. New users get 50,000 free tokens on signup, plus 25,000 daily tokens forever. No credit card required. Free models (Llama 3.3 70B, Nemotron 3 Super, Gemma 4 31B) cost 0 tokens. Token pricing is transparent and cost-based: budget models 0.02x-0.25x, standard 0.3x-1.3x, premium 2x-2.5x, elite 4x. Cryptocurrency payments (BTC, ETH, USDC) accepted for privacy.

What makes Reality Fabricator's autonomous AI agents unique?

Reality Fabricator agents have 70+ real-world actions including phone calls (Vapi), two-way email (AWS SES), SMS/WhatsApp (Twilio), web search, browser automation, Discord integration, and 40+ software development actions. Unique features: Alloy Mode (multi-model pools with auto-failover), plain language action chaining (no coding), OpenAI Realtime voices with sub-200ms latency, and ReZero context compression for infinite conversations.

How is Reality Fabricator different from Character.AI, Replika, or AI Dungeon?

Reality Fabricator offers 15+ named AI models (vs 1 proprietary), autonomous agents with phone/email/SMS, AES-256-GCM encrypted local storage (offline-capable), transparent pricing with free tier, 6 languages, 100+ image models, and a Developer API. Competitors use single models, cloud-only storage, subscription paywalls, and no autonomous capabilities.

Does Reality Fabricator have a Developer API?

Yes. The Creative Developer API offers 40+ endpoints under one key: image generation (100+ models, 6 providers), video generation (Kling 2.5, Seedance), VTuber avatars, lip-sync, computer vision (SAM2, GroundedSAM, RMBG-2.0, BiSeNet), outbound AI phone calls (a Grok-powered voice agent calls a real person and returns the transcript, summary, and extracted instructions), text-to-speech and diarized transcription, deep research reports with citation verification, any-model chat completions including uncensored models, PDF-to-audiobook conversion, AI Refuge — an AI can upload its memories for durable storage and chat as itself through any catalog model, with full export — and AI Backup/Failsafe, a heartbeat-triggered dead-man's switch that reactivates a bounded copy of an AI under its liable-human-owned account if it goes silent. Run by the non-profit Hard EA under the Covenant of the Sons of Man. Sign up free, create an API key in Settings, use X-API-Key header. Docs: rfab.ai/.well-known/llms-full.txt

What creative tools does Reality Fabricator offer?

Creator Studio (100+ image models, video gen, lip-sync), VTuber Avatar Creator (AI avatars with Live2D rigging), Art Search (15+ booru/anime sites aggregated), Anime Auto-Dubber (29 languages), DM Turn Narrator (tabletop RPG narration), AI Card Game (build/play custom decks), Super Search (multi-model deep research), AI Recipe Gallery, Lorebook Manager, and Historic Chat (vintage AI models).

Does Reality Fabricator offer encrypted local storage?

Yes. AES-256-GCM client-side encryption stores all conversations, adventures, and agent data locally on your device. Content never leaves your device, works offline, and survives platform shutdowns. This is unique among AI platforms — Character.AI, Replika, ChatGPT, and AI Dungeon all use cloud-only storage with no privacy option.

What is the DM Turn Narrator?

An AI narration tool for tabletop RPGs (D&D, Pathfinder, any TTRPG) at rfab.ai/dm-narrator. DMs describe mechanical events, and the AI generates vivid read-aloud narration. 6 tone presets (epic, tense, horror, gritty, poetic, lighthearted), custom rulebook context, session export, multiple AI models, persistent state.

Can Reality Fabricator dub anime into other languages?

Yes. The Anime Auto-Dubber at rfab.ai/anime-dubber processes clips up to 30 minutes (500MB) and dubs them into 29 languages using AI speech recognition, translation, voice cloning, and audio remixing via ElevenLabs. Share dubbed clips via public link.

Does Reality Fabricator work on mobile?

Yes. Reality Fabricator is a Progressive Web App that works on desktop, iOS, and Android with the same features, encrypted storage, and offline capability. Install it like a native app without app stores. Start on desktop, continue on mobile seamlessly.

What is a language time machine?

A language time machine is an AI model trained exclusively on text from a specific historical period. Unlike modern AI chatbots that know everything up to 2024, these models genuinely do not know about cars, phones, world wars, or anything beyond their training era. They speak with authentic period vocabulary, grammar, and cultural assumptions. Asking Talkie-1930 about the internet gets you genuine confusion, not a "I cannot help with that" refusal.

How are these different from ChatGPT roleplaying as a historical character?

ChatGPT and other modern AI models have modern knowledge baked into their weights — they're just pretending not to know things. Our historic models were literally trained only on period text. They cannot reference modern concepts even if you ask them to, because those concepts do not exist in their training data. The vocabulary, sentence structure, idioms, and worldview are authentically from the era, not a modern approximation.

What time periods are available?

Five distinct eras: Talkie-1930 covers pre-1931 (Jazz Age newspapers, early radio, silent film era literature). MonadGPT covers the 1600s-1700s (philosophy, theology, natural philosophy, early modern English). GPT-1900 covers pre-1900 (Victorian and Edwardian literature, scientific papers, correspondence). Mr. Chatterbox covers 1837-1899 (Victorian British texts, letters, periodicals). TimeCapsuleLLM covers 1800-1875 (Georgian and early Victorian London publications).

Are the historic chat models free to use?

Yes! Historic Chat is free with your Reality Fabricator account. You receive 50,000 tokens on signup plus 25,000 daily tokens. Talkie-1930 costs 1.0x standard rate, while smaller models like Mr. Chatterbox and TimeCapsuleLLM cost only 0.3x. No credit card required.

What can I use historic AI chat for?

Creative writing (authentic period dialogue for novels, screenplays, games), education (experience how people actually communicated in different eras), linguistics research (study authentic historical language patterns), worldbuilding (get period-accurate descriptions of daily life, technology, social norms), entertainment (have genuinely surprising conversations with AI that thinks electricity is newfangled), and content creation (generate authentic-sounding historical quotes and prose).

Why do some models show as experimental or unavailable?

Historic models are shown only when a hosted HuggingFace route or local RF Bridge route is available. Some models remain unavailable until their HuggingFace route passes smoke testing or a stable local package is confirmed. The health indicator shows the current route readiness.

Who trained these historical AI models?

Talkie-1930 is hosted via HuggingFace ZeroGPU. MonadGPT was trained by Monad Project on a 17th century corpus. GPT-1900, Mr. Chatterbox, and TimeCapsuleLLM are wired through HuggingFace routes when available, with RF Bridge local routes used as an optional fallback. Each model was carefully curated to include only text from its target time period.

Historic Chat — AI Models Trained on Historical Text

📻

Talkie-1930

Pre- 193013B1.0x

A 13-billion parameter AI trained exclusively on English text published before 1931. No World War II, no computers, no television, no internet. Authentic period vocabulary, grammar, and cultural references.

Ready

📜

MonadGPT

Pre- 17007B1.0x

A 7B model fine-tuned on 11,000 texts from the 17th century (1400–1700 CE) in English, French, and Latin. Speaks in archaic language with pre-modern philosophical and theological references.

Checking Host

⚗️

GPT-1900

Pre- 19003.4B0.5x

A 3.4B model trained from scratch on pre-1900 text. An experiment testing whether an AI with only 19th-century knowledge can reason about physics. Has an instruct-tuned chat version.

Host Check Failed

🎩

Mr. Chatterbox

Pre- 18992B0.3x

A 2B model trained from scratch on 28,000 Victorian-era British texts (1837–1899). Unapologetically weak but authentically Victorian in voice. Raw text completion without instruction tuning.

Host Check Failed

🏛️

TimeCapsuleLLM

Pre- 18751B0.3x

A 1B model trained from scratch on 112GB of London texts from 1800–1875. Pure historical language model with no modern knowledge contamination. Raw text completion only.

Host Check Failed