FantasyLab.ai · local-first AI tools for serious work

Glass-box AI.
Local. Open. Yours.

Three tools, one philosophy: AI that shows its work and runs on your machine.
Aurora computes cited findings from your data. Studio turns a sentence into a playable 3D game or film. EmberOS makes your phone a live terminal for your own computer.

// real math · // real physics · // real cinema · // real hardware

Aurora → Fantasy Studio → EmberOS →
scroll — fly through the lab
// the philosophy

Black-box AI is fine for chat.
Not for work that matters.

Cloud AI tools generate confident answers and hide their math. FantasyLab builds the opposite — glass-box tools that run on your machine, show every step, and produce artifacts you can audit, version, and own.

🔍 Glass-box at every layer 💻 Local-first execution 📖 Source-available · Apache 2.0 + BSL 1.1 🎯 Honesty over impression
// station 01 · aurora · quantitative intelligence

Computes,
not guesses.

Drop a CSV → cited findings, anomalies, forecasts, and the governing equation — in seconds, entirely on your machine.

Explore Aurora → ▶ Try the demo — no install
v2.0 · 599 tests · 🥇 #1 Product of the Day
// station 02 · fantasy studio · playable games + cinema

Describe it.
Play it or film it.

Type a sentence → a playable 3D game with real physics, or a cinematic film rendered in Blender. No code, no cloud, yours forever.

Explore Studio → View on GitHub →
v1.0 · live now · source-available
Aurora · Quantitative

If a model can't show its work,
it shouldn't make claims.

Most AI analytics tools are LLMs reading your spreadsheet and pattern-matching plausible-sounding sentences. They hallucinate. They can't be reproduced. They can't be audited.

Aurora runs real research-grade math first — SINDy, HMM, mutual info, persistent homology — and only then asks a local LLM to translate the findings, with every claim grounded in cited public knowledge.

Studio · Cinematic

If a renderer can't simulate light,
it isn't really 3D.

Most "AI video" tools are diffusion models that hallucinate pixels frame-by-frame. They can't render the same shot twice. They can't be edited. They have no source file.

Studio uses Blender's Cycles engine — real path-traced light simulation — directed by a local LLM that plans camera, lighting, and atmosphere. Every render is a real .blend file you keep forever.

Different domains. Same engineering principle.
Run the actual physics. Trust the math. Show your work.

Three products. One lab.

Built independently, shipped together, sharing the same local-first foundation. Pick the tool that fits your work — or use all three.

AURORA
v2.0 · 599 tests · 🥇 #1 Product of the Day
Local AI that computes — not guesses.
For humans analyzing hard data. For AI systems that can't afford to hallucinate. Cloud LLMs guess. Aurora computes.
🧠 Aurora Copilot
for analysts, quants, scientists, engineers

Drop a dataset and get rigorous findings — anomalies surfaced, causal relationships tested, forecasts with confidence bounds, every claim cited to the underlying computation. No cloud LLM guessing. No black-box math.

  • 24+ research-grade methods (Isolation Forest, Granger, HMM, SINDy, persistent homology, Pearl do-calculus, VAR, DTW, BOCPD, Robust PCA, Kalman, EMD)
  • Six analytical lenses + spacetime system graph + phase-space trajectory
  • Knowledge-grounded synthesis · seed:* citations · custom KB ingestion (PDF/MD/TXT → bank)
  • v2.0 just shipped: causal inference, multi-dataset joins, composable findings, Plugin SDK
🛡️ Aurora Cortex
for AI builders, agent developers, AI product teams

The verification layer your AI agents and AI products call when they can't afford to hallucinate quantitative claims. Connect via MCP, the Python SDK, or Decision Contracts.

  • Aurora MCP server — 7 tools, HTTP transport (Claude Desktop, Claude Code, Cursor, custom agents)
  • Python SDK with .aurora.json bundle format (SHA-256 + Ed25519 + signer registry)
  • Decision Contracts — 6 action types (webhook, Slack, Discord, email, log, file) · SSRF-guarded
  • BYO-LLM — Anthropic, OpenAI, Gemini, Ollama, OpenAI-compatible · GPU embeddings (CUDA/MPS)
  • Streaming connectors — file-watcher, SSE bus, Kafka, Postgres CDC
Aurora Studio with Anomalies lens + intelligence tiles [ aurora console — anomalies lens + intelligence tiles ]

A real run on a 9,357-row air-quality dataset. AUTO tier. ~14 seconds local. 0 fabricated. 12 cited knowledge entries.

599tests passing
24+research-grade methods
6analytical lenses
100%local execution
Get Aurora → View on GitHub
// try it now — no install · or download the desktop app
// or the developer path
git clone https://github.com/FantasyLab-ai/aurora.git
cd aurora
python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements.txt

# Optional substrate-layer extras
pip install cryptography           # Ed25519 bundle signing
pip install mcp                    # MCP server for LLM agents

# Run the Studio
python studio_api.py

Open http://127.0.0.1:8000 → click ▶ Try a demo → 10-second smoke test, or drop your own CSV / Parquet / JSON / XLSX.

// award-winning on PeerPush
🥇 #1 Product of the Day on PeerPush →
STUDIO
v1.0 · live now
Local AI that directs real 3D cinematic video from a single prompt.
Type a scene. AI directs casting, camera, lighting, and pacing. Blender renders it locally with Cycles ray-tracing. Real light. Real .blend files. No cloud, no credits, no usage limits — yours forever.
Fantasy Studio canvas with a cinematic render in progress [ studio canvas — cinematic render ]
  • Cycles ray-traced rendering · real photons, not pixels
  • Local Gemma 3 12B director · 4 render tiers
  • 15 cinematic recipes · 300+ curated assets
  • Refine with words · iterate scenes by chatting
  • MP4 + GIF + PNG sequence + .blend source export
Get Studio → View on GitHub
EMBEROS
new · live now · source-available
A real computer in your pocket — your phone is the Head, your machine is the Engine.
EmberOS turns any phone into a live terminal for the computer you already own. A tiny daemon (emberd, one static Go binary) runs on your desktop or server; your phone pairs to it over your own network and gets a whole desktop — no cloud in the middle, no accounts, no telemetry. Everything Ember can see or do lives on hardware you own.
  • Real filesystem, real PTY terminal, real code editor with a coding agent
  • ✨ Ember assistant with vision — local models via Ollama, or your own Claude / GPT key
  • Game Boy → Genesis arcade with battery saves synced to your Engine
  • Agent Inbox: AI proposes, you approve — every action audited & reversible
  • Keys and tokens never leave the Engine (0600 files, per-Head revocable pairing)
// twelve live surfaces in the current build
FilesTerminalEditor + agent Arcade · 5 consolesModels · vision✨ Ember InboxGitHubForge MailCalendarAutomations
It completes the lab: Aurora is the analyst, Studio is the filmmaker, and EmberOS is the computer that carries them in your pocket — same glass-box promise, same local-first foundation.
Get EmberOS → View on GitHub

What every FantasyLab tool shares.

All three products inherit the same foundation. These are the principles that don't bend.

💻

Local-First

Your CPU. Your GPU. Your data. Zero cloud dependencies. No API costs. No tokens. No credits. The whole pipeline runs on your machine.

🔍

Glass-Box

Every decision inspectable. Every method traceable. Every output reproducible. No mystery layers. No black boxes. No "trust us."

⚖️

Anti-Hallucination

Math first. Physics first. LLMs only translate what's already been computed or rendered — never invent. Verifiers catch what slips through.

📜

Source-Available

Aurora: Apache 2.0. Studio: BSL 1.1. Read the code. Audit the methods. Run it yourself. Modify it for your work. The opposite of a black-box SaaS.

// the manifesto

We think AI got the trade-off backwards.

The industry decided that "intelligent" means a model that confidently generates anything, instantly, without showing how. Faster, fancier, more impressive — and quietly, more wrong.

We think a tool that admits what it doesn't know, runs on your machine, cites every claim, and produces output you can audit for the rest of your life is more valuable than a cloud subscription that hallucinates beautifully.

That's the FantasyLab.ai bet. Local-first, glass-box, math-grounded, source-available tools — built for serious work, by serious builders who refuse to ship vibes as if they were truth.

// shipped from a single workstation, in public, without venture money

// built in public
GitHub Org
FantasyLab-ai
studio · aurora · public
Aurora · Knowledge Entries
52K
growing nightly · target 2–3M
Studio · Curated Assets
300+
healed · ground-aligned · ready
License
Apache 2.0 · BSL 1.1
aurora · studio · source-available forever

Why FantasyLab.

The questions we get asked most. Honest answers.

Why two products under one roof?

Because the same engineering philosophy — refuse the hallucination shortcut, run the actual physics, ship local — applies to creative work and analytical work. Studio and Aurora are different tools solving different problems with the same conviction.

Why local-first?

Privacy, cost, control, and reproducibility. Your data never leaves your machine. There's no monthly bill that creeps. There's no rate limit on a deadline. And a year from now, the same input still produces the same output.

Are these toys, or production tools?

Production tools, with rough edges. Studio renders real cinematic video. Aurora finds real anomalies and fits real ODEs. Both are early — we ship in public and iterate fast. Quality compounds.

How is this funded?

Patreon, donations, and eventual premium tiers on top of the open-source core. No venture money, no ads, no data harvesting, no enterprise sales cycle. Just builders, supporters, and tools.