impforge Svelte Themes

Impforge

ImpForge — your desktop, your AI, your rules · Local-first AI workstation that builds anything (SaaS, mobile, trading bots) with built-in compliance · Pop-out preview windows · Paper-trading-first · 100% offline · Zero data collection · Free forever tier + EUR 25/mo Pro · Free MIT template gallery on GitHub · Made in Germany

ImpForge

ImpForge — DigiU

Your Desktop Becomes an AI Command Center

A floating AI companion that turns any laptop into a sovereign workstation
with 480+ modules. Runs 100% offline. Sees no data. Tracks nothing. Ever.

🆕 April 2026 update: ai_imp now builds anything for you — describe what you want in plain English, watch a working app pop out in its own window seconds later. Production-grade SaaS with built-in compliance. Mobile apps. Trading companions that start in paper mode and only graduate to real money after they've proven themselves. All on your laptop. Whether you want to peek under the hood or not, our template gallery is now MIT-licensed on GitHub — fully open, free to use anywhere, even with Cursor, Bolt, or Claude Code.

Templates on GitHub Skills on GitHub template.json v1 spec

HyperChatThe AIAnti-Palantir Shield4-Model AIYour DesktopPrivacy PromisePricingGet Started

Offline AI Anti-Palantir Shield EU AI Act Compliant AES-256 + ZK Proofs Zero Data Collection No Tracking Platform Free Made in Germany


The problem with every "AI workstation" you've tried

Your data leaves your machine. The AI is lobotomised. The interface is a sidebar stuck to the edge of your IDE. You pay a cloud bill that grows every month. The "assistant" stalls on sensitive topics and forgets what you told it last week. Your tools are scattered across 14 browser tabs, 3 apps, and a Discord server.

You wanted a command center. You got a subscription to someone else's surveillance system.

ImpForge is the alternative.


HyperChat — The Heart of ImpForge

Forget app windows that eat your screen. ImpForge lives in HyperChat — a small, beautiful AI panel that floats at the edge of your display.

It's tiny. It's smart. It moves on its own.

  • Magnetically docks to any screen edge — left, right, top, bottom
  • Automatically moves out of the way when you're working
  • Grows when you need it, shrinks when you don't
  • Everything starts here: just talk to your AI

Think of HyperChat as a portal. You tell it what you need, and the right tools appear on your desktop as floating windows. An IDE here. A spreadsheet there. A chart in the corner. You arrange them however you want.

HyperChat is how you command your entire desktop.

It shows you what your AI is thinking. It shows you your goals, deadlines, and predictions. It streams every LLM token, every tool call, every state change through a live event protocol so you see the AI working — no black box, no "loading spinner theatre". It lets you switch between 5 levels of AI autonomy — from "only answer when asked" to "handle everything yourself" — each level guarded by a cryptographic audit chain that logs every decision.

No other tool works like this. Most AI assistants are a sidebar. HyperChat is a command center.

AI-Readable Design System — The world's first UI that evolves itself.

ImpForge has a built-in design intelligence system that no other AI tool has:

  • Universal UI Scanner — reads any UI component and checks it against 400+ design tokens for WCAG accessibility, font correctness, spacing consistency, and animation quality. Generates a score and concrete improvement suggestions.
  • AI-Readable Design Tokens — every color, font, spacing value, animation, and effect is defined in a machine-readable format that the AI understands natively. No guessing. No generic output.
  • Google A2UI v0.9 Support — ai_imp can describe entire UIs in a flat, declarative JSON format. 20+ trusted component types in a security-sandboxed catalog. The AI generates UI, your app renders it — safely.
  • W3C Design Tokens Standard — export tokens to the international standard (W3C DTCG, Oct 2025). Compatible with Figma, Tokens Studio, Style Dictionary, and any design tool.
  • Works with every AI tool — Claude Code, Cursor, Windsurf, JetBrains AI, Copilot — they all read the same design rules automatically.

An AI that actually gets you

It remembers. Actually remembers.

Not "we saved your chat history." Real memory. ImpForge builds a picture of who you are over time — your skills, your preferences, your projects, your goals. Every conversation makes it smarter about you specifically.

Ask it something on Monday. Reference it on Friday. It connects the dots.

It thinks before it speaks.

Most AIs just throw words at you. ImpForge's AI has an actual thought process. When you ask a hard question, it checks your code, looks at your past decisions, considers what worked before, and then answers.

You can watch it think through HyperChat's Thinking Panel — see which parts of its brain light up while it works.

It has 472 personalities. And picks the right one.

Working on a React component? It becomes a frontend expert. Writing a business proposal? Switches to a business consultant. Doing your taxes? Becomes a German tax advisor who knows the current deadlines.

You don't configure this. It just happens.

It has a board of directors.

For the big decisions, your AI assembles a virtual boardroom: a CTO, a CFO, a legal advisor, a security expert — 10 specialists who debate your question from different angles. They argue. They vote (weighted by a Hebbian trust score that learns which advisor has been right in your context). Then they give you a recommendation with dissenting opinions clearly noted.

Like having a team of advisors who never sleep and never charge by the hour.

It scales from helper to autonomous.

Three operating tiers, one kill-switch. Supervised (the AI asks before every action). Guarded (auto-runs safe actions, asks for sensitive ones). Autonomous (full self-governance for trusted long-running jobs). Every decision at every tier is recorded in a SHA-256 Merkle audit chain — tamper-evident, mathematically verifiable, yours alone.

It gets better every day.

The more you work with ImpForge, the more it learns. After a week, it anticipates what you need. After a month, it feels like it reads your mind.

All that learning stays on your machine. Delete the app, it's gone. No traces anywhere.


Anti-Palantir Privacy Shield

"Your AI assistant that even Palantir can't spy on."

Data surveillance platforms like Palantir work by fusing data from multiple sources into searchable profiles. They need three things to exfiltrate data (the "Lethal Trifecta"): untrusted input, access to sensitive data, and the ability to communicate externally.

ImpForge breaks all three.

5-Tier DigiU Privacy System

Choose your protection level during setup. It controls EVERYTHING — AI routing, storage, network access, telemetry, and sharing.

Tier Name What It Means
0 Fortress Air-gapped. Zero network. Military grade. Your data physically cannot leave your device.
1 Vault Local-only processing. Encrypted storage. No cloud sync. Enterprise GDPR strict.
2 Shield Local AI primary. Anonymous cloud fallback with differential privacy. No PII ever sent.
3 Bridge Encrypted cloud sync. E2E encryption. P2P sharing with colleagues.
4 Open Full cloud integration. Maximum productivity. Privacy still protected by encryption.

5 Defense Layers (Always Active)

  1. PII Detection — Scans every outbound message for 13 types of personal data (emails, SSNs, credit cards, API keys, passwords, phone numbers, etc.) and auto-scrubs before any cloud call
  2. Prompt Injection Firewall — Blocks 17 attack patterns including instruction override, data exfiltration, and hidden instructions in HTML/Unicode
  3. Canary Tokens — Invisible markers in your documents. If your data appears in any external database, you'll know.
  4. Cryptographic Audit Trail — Blockchain-style Merkle tree logs every privacy-sensitive operation. Tamper-proof. Mathematically verifiable.
  5. Governance Graph — 23 rules control which AI agents can access what data. Deny-by-default. External tools need explicit permission.

EU AI Act — Already Compliant

While competitors scramble to comply with the August 2026 deadline, ImpForge ships compliant by default:

  • All AI features classified by risk tier (Minimal, Limited, High-Risk)
  • Transparency labels auto-generated for AI content
  • Full compliance report at the click of a button
  • GDPR Art. 25 (Privacy by Design) baked into the architecture

Zero-Knowledge Proofs + Encrypted Computation

ImpForge can mathematically prove that your data was processed correctly — without revealing what the data is:

  • SHA-256 Commitment Schemes — prove data existed at a point in time
  • AES-256-GCM Encrypted Queries — cloud APIs never see your plaintext
  • HMAC Verification — prove computation integrity without revealing inputs
  • Tamper-Evident Audit Logs — any modification is mathematically detectable

Every line of ImpForge code was written assuming someone hostile is watching. Because they are.


Your desktop, your rules

400+ tools, one HyperChat command away.

AI Chat. Code Editor. Document Writer. Spreadsheets. Email. Calendar. Notes. PDF Viewer. Docker Manager. GitHub Dashboard. Trading View. Browser. Terminal. CRM. Analytics. Presentation Builder. Campaign Manager. Multi-repo orchestrator. Live-broker dashboard (Alpaca, IBKR, Binance, Kraken, Coinbase). And 380 more.

Each one is a building block. Tell HyperChat to open it — it appears as a floating window on your desktop. Drag it anywhere. Resize it. Pin it. Make it transparent. Build the workstation you've always wanted.

Builders, not templates.

ImpForge doesn't give you "sample projects." It gives you builders. Ask HyperChat for a SaaS, a mobile app, a desktop app, a CRM, an e-commerce store, a firmware project, or a single-page site — and the AI scaffolds the whole thing, runs it in an isolated sandbox, and iteratively improves it with quality feedback until it works. Twelve categories. One command each.

Every app appears in its own window.

No iframe juggling. No "deploy to see it." When ai_imp finishes building your app, a fresh window pops out on your desktop running the live preview — hot-reloaded, interactive, yours to click through. Move it to a second monitor. Leave it open while you describe the next change. It updates as you speak.

Compliance comes built-in.

Building a fintech app? ai_imp ships FINRA-aware audit logging by default. Healthcare? HIPAA-ready schema. SaaS platform? Multi-tenant isolation auditing catches the one-line bug that causes 92 % of SaaS data breaches — before your migration ever hits the database.

We put the scaffolding blueprints on GitHub under an MIT license so anyone — even people who'll never buy ImpForge — can read, fork, or redistribute them.

Use them standalone with any tool you like. Or let ImpForge's engine turn them into live apps in one conversation. Your call, both work.

Trading that earns the right to trade.

ai_imp can follow your investment strategies — but no agent gets real money until it has proven itself. Every trading companion starts in paper mode. It stays there until it's racked up 200 winning sessions, with zero regressions, in market conditions below the VIX breaker threshold. Only then does promotion to real-money micro-trading unlock. You never wake up to a surprise.

Workspaces — Your personal command centers.

Create a "Coding" workspace: IDE on the left, terminal at the bottom, AI chat on the right. Create a "Trading" workspace: charts everywhere, news feed, portfolio tracker. Switch between them in one click.

Each workspace remembers everything — window positions, AI context, active tools, even which personality your AI was using.

It's like having multiple desktops on steroids. The kind of setup you've only seen in movies or on trading floors that cost EUR 10,000+. Except it's on your laptop, and the base version is free.


Our Privacy Promise

This isn't a section we added because lawyers told us to. This is why we built ImpForge.

We collect NOTHING.

Not analytics. Not usage data. Not error reports. Not "anonymized telemetry." Nothing. Zero. Null.

When you use ImpForge, no packet of data leaves your computer heading to our servers. We don't have servers that receive user data. We didn't build them. We don't want them.

We sell NOTHING.

Your conversations with the AI? Yours. Your documents? Yours. Your browsing habits? We don't even know them. There is no business model that involves your data. Our business model is: we make good software, you pay EUR 25/month if you want the pro features. That's it.

We track NOTHING.

No cookies. No analytics scripts. No "we use cookies to improve your experience." No fingerprinting. No device IDs. The app doesn't phone home. Not on launch. Not ever.

The AI runs on YOUR machine.

ImpForge's AI doesn't send your questions to a cloud API. It runs locally, on your hardware, using open-source AI models. Your conversations never touch the internet.

Military-grade encryption protects your data.

Every database on your machine is encrypted with AES-256 — the same standard used by governments and militaries. Protected by a separate password that even the AI cannot enter. Even if someone steals your laptop, they get encrypted noise.

Even other AI tools can't see it.

Using Cursor, Copilot, or Claude Code alongside ImpForge? They can read files on your disk — but ImpForge's databases look like random gibberish to them. Encrypted at rest. Always.

You're in 100% control. Always.

  • Want to delete everything? One click. Gone. All data erased.
  • Want to export your data? It's yours. Take it anywhere.
  • Want to see what ImpForge knows about you? There's a dashboard for that.
  • Want to use ImpForge without creating an account? There is no account. There never will be.

This is not marketing. This is architecture.

We didn't add privacy features after building the product. We built the product around privacy. The encryption, the local processing, the zero-collection policy — these aren't toggles you can turn off. They're how the software works at its core.

AiImp Technology's commitment: We will never collect, store, sell, share, or analyze any user data. Not today. Not when we're bigger. Not when investors ask us to. This is a founding principle, not a business decision.


Simple pricing. No tricks.

Free Pro (EUR 25/mo) Founder's Edition (EUR 750 one-time)
HyperChat AI Panel Yes Yes Yes
All 400+ modules Yes Yes Yes
Local AI (runs on your PC) Yes Yes Yes
Encrypted databases Yes Yes Yes
AI personalities 50 472 472
Quality rules 50K 158K+ 158K+
Board of Directors Yes Yes
Digital memory Basic Full Full
Enterprise features Yes Yes
Workspaces 2 Unlimited Unlimited
Live broker integrations 5 providers + simulator 5 providers + simulator
Third-party providers 12 across 6 sub-namespaces 12 across 6 sub-namespaces
Autonomy tiers Supervised All 3 (Supervised / Guarded / Autonomous) All 3
Updates Community Priority 3 years included

Free forever. No credit card. No trial. The free version is a real AI workstation — not a demo.

Founder's Edition: One-time EUR 750. Limited to first 500 licenses. App works forever. When the 500th license sells, the price goes up — it will never drop.

Trust, with receipts

We don't just say "we care about quality." We measure it. Every number below is re-checked before every release.

Claim Measurement Value
Zero crashes from panics .unwrap() that can actually panic in the production Rust tree 0
Zero technical debt hiding #[allow(dead_code)] attributes anywhere in 572 files 0
Zero compiler warnings cargo check across all 4 build modes (default, --features swarmforge, --tests, --tests --features swarmforge) 0
Zero build drift Rust toolchain pinned via rust-toolchain.toml 1.95.0
Zero data leaks Lines of code sending your data to us or anyone else 0
Every module stays healthy Neural Trias (Memory + IPC + Health) covers every module 313/313
Every sub-module is emergent Sub-modules declared via the Emergence Kernel 82+
Every gateway decision is logged SHA-256 Merkle chain in the Autonomy gateway every one
Every commit is auditable Commits signed + reproducible (public release tags) every one

These aren't badges we bought. They are the output of cargo check, grep, and git log on the public build — you can re-run any of them yourself.


4 AI Models. 7.7 GB. Runs on any laptop.

Most AI tools need expensive cloud APIs or 32 GB of RAM. ImpForge runs 4 specialized AI models on your machine — even an 8 GB laptop.

Model Role What it does
qwen3-imp:8b The Brain Deep reasoning, strategic decisions, code generation. Selectively aligned (helps everywhere professional, refuses only violence/terrorism/fraud/exploitation). Has a "thinking mode" for hard problems.
Qwen2.5-Coder 1.5B Fast Hands Code completion in under 100ms. Fill-in-the-middle, syntax checking, refactoring.
nomic-embed-text Memory Turns your documents into searchable vectors. Powers the knowledge base and semantic search.
SmolLM2 1.7B Classifier Understands what you want in under 50ms. Routes your request to the right tool.

Smart cascade routing: 60-70% of requests are handled by the tiny Classifier (instant). Only complex questions reach the Brain. This saves 3-5x compute.

Self-learning routing: The system tracks which model performs best for each type of task. After just 5 interactions, it automatically routes to the optimal model. It literally gets smarter the more you use it.

12-step quality pipeline: Every single AI response passes through privacy scanning, governance checks, quality validation, self-correction, PII scrubbing, cryptographic sealing, stigmergy trace, session context, evolution record, EU-compliance labelling, and module-memory feedback — automatically, invisibly, every time.

Board of Directors: For big decisions, the Brain simulates 10 executives (CEO, CTO, CFO, CMO, COO, CIO, CLO, CHRO, Analyst, Auditor) — each with a different perspective and temperature. They vote. You get a recommendation with dissenting opinions noted.

Works with every AI tool: 227 MCP tools organised into 13 namespace micro-servers (90% token savings vs monolithic). Claude Code, Cursor, JetBrains, Copilot, Windsurf — they all talk to ai_imp like a brother. Lazy schema loading: tool names first (~2K tokens), full schema on demand.

Selective Alignment: Your AI is helpful and safe. We don't ship a lobotomized model that refuses to write a port scanner. qwen3-imp helps with security research, creative writing, medical info, legal analysis — everything professional. But it firmly refuses violence, terrorism, fraud, and exploitation. No wholesale censorship. No wholesale uncensoring. Enterprise-grade balance.


Get started in 2 minutes

What you need

  • A computer with 8 GB RAM (16 GB recommended)
  • Linux, Windows 11, or macOS
  • No API keys. No accounts. No cloud setup.

Install

# Linux
wget https://github.com/AiImpDevelopment/impforge/releases/latest/download/ImpForge.AppImage
chmod +x ImpForge.AppImage
./ImpForge.AppImage

# Windows — download ImpForge-setup.exe from Releases
# macOS — download ImpForge.dmg from Releases

What happens next

  1. HyperChat appears — a small AI panel at the edge of your screen
  2. Your AI introduces itself: "Hi! I'm your ai_imp. Let's set up your workspace."
  3. It scans your hardware and recommends the best local AI model
  4. You start talking. It starts learning. It starts helping.
  5. Welcome to your new desktop.

Built in Germany. For everyone.

ImpForge is made by AiImp Technology in Germany.

We believe AI should be:

  • For everyone — the free version is real software, not a teaser
  • Private — your data never leaves your device, period
  • Offline — works without internet, works on old hardware
  • Honest — no dark patterns, no surprise charges, no data harvesting

Not like Grok. Like DeepSeek — open, democratic, accessible.

"The best AI is the one that respects you."

By the numbers

Lines of code 730,000+
Automated tests 6,742+
Rust modules 645 files, 322 modules
Sub-module files (focused micro-programs) 300+ across 30+ directories
Free template gallery (MIT) 78 industry × category templates — rolling out in 5 waves
Free skill library (MIT) growing catalog of reusable AI agent skills
template.json v1 specification published on GitHub — adopted by ImpForge, open to all
AI personalities 472
Quality rules 158,000+ (81 AI-generated at benchmark 1.00)
Tauri commands 2,975
MCP tools (Claude Code / Cursor / JetBrains / Copilot / Windsurf) 227 in 13 namespace servers
MCP token savings vs monolithic 90% (lazy schema loading)
4-Model AI Pipeline benchmark 1.00/1.00 (all 6 dimensions perfect)
qwen3-imp:8b Selectively aligned brain — PRODUCTION READY
Self-learning routing Adapts from quality feedback automatically
Memory-informed routing Module memory overrides classifier when quality > 0.85
Neural Trias Module Memory + Module IPC + Module Health (self-healing)
Emergence Kernel One call wires every module's memory + IPC + health + MCP + insight bus
Lazy LSP Sidecars 12 languages load only when you code them — Pyright's 1.7 GB only appears when you open a .py
Pipeline Modes 7 (Streaming, Thinking, LongForm, Secure, Diverse, Code, Standard)
Widget Isolation 10 built-in widgets, <5MB idle, ~1MB suspended
Project Builders 12 categories (SaaS, Web, Mobile, Desktop, Firmware, CRM, E-commerce, Single-Page + sandbox + iterate + helpers + types)
Live broker integrations 5 (Alpaca, IBKR, Binance, Kraken, Coinbase) + deterministic simulator
Third-party providers 12 across 6 sub-namespaces (finance, content, cloud, payments, comms, social)
Autonomy tiers 3 (Supervised / Guarded / Autonomous) with SHA-256 Merkle audit chain + fleet-wide kill switch
HyperChat-Next events 11 AG-UI event variants (stream every LLM token + tool call + state change)
Autonomous micro-program modules 400+ (each self-sufficient)
Module capabilities (each) Privacy + Governance + Context + Stigmergy + Evolution + Memory
Anti-Palantir defense layers 5
DigiU privacy tiers 5 (Fortress to Open)
Selective Alignment Engine 8 safety categories REFUSED, 10 alignment categories HELPED
Neural synapses 175 (40 in Synapse Fabric v2)
Specialized AI agents 15 (Privacy, Compliance, Security, Governance, Evolution, ...)
Resource power modes 5 (DeepSleep <80MB to Genesis 5GB VRAM)
Board of Directors 10 executives via temperature variation + Hebbian trust voting
Governance rules 23 (deny-by-default)
Emergent intelligence layers 4 (Pattern, Anomaly, CrossModule, Summary)
ai_imp action types 25
A2UI components 20+
Design tokens 400+
Research papers cited 608+
EU AI Act features classified 17 (all compliant)
Build errors 0
Build warnings 0
Production .unwrap() 0 (Crown-Jewel grade)
#[allow(dead_code)] 0 anywhere
Rust toolchain Pinned 1.95.0 (reproducible builds)
Data collected from you 0

ImpForge — DigiU
Your desktop. Your AI. Your rules.

DownloadFeedbackBuild from Source

Top categories

Loading Svelte Themes