The BYOB of AI: bring your own model, backend, cloud key, or local runtime. More AI keeps it governed, traceable, and under your control.
More AI is a free, open-source Windows desktop application for people who want one serious AI workspace without surrendering choice. Chat with cloud APIs, run local LLMs, code with tool agents, create music, images, video, documents, and workflows, orchestrate swarms, and audit what happened afterward. Every execution runs through a governance engine so nothing happens silently. No subscriptions. No API surprises. Run fully local, use cloud APIs, or mix both — you decide.
The goal is simple: make powerful AI feel usable, inspectable, and yours — no command-line setup required. Download the installer, run it, and the app handles everything else. The built-in terminal is there when you want it, but you should not need one just to get started.
Built by Doug Trier / Trier OS · Open source under the MIT License
More AI started as the AI workspace I wanted to move into: local-first, governed, traceable, and deterministic where it matters. I wanted agents that could help me build, create, research, and automate without silently running wild or locking me into one provider.
More AI is intentionally agent-agnostic. Use Ollama, llama.cpp, BitNet, Hugging Face models, LM Studio, OpenRouter, OpenAI, DeepSeek, Anthropic, Google, or your own OpenAI-compatible endpoint. The goal is simple: you choose the model, you choose the runtime, and More AI keeps the workflow visible, gated, and under your control.
I am sharing it because other people may want the same kind of workspace. If you want to help make it better, test it on your hardware, report bugs, improve docs, contribute code, or help shape the roadmap. Let's build a practical local-first AI community around tools people can actually use.
This is an actively developed project. Core workspaces — Chat, IDE, Music Studio, Picture Editor, Video Editor, and AI Colosseum — are solid and used every day. Some advanced features (P2P Collaborate, Voice Studio, AI WebView co-navigation) are still being refined and may have rough edges. When you find something broken, open an issue — every report is read and taken seriously. Every fix makes the app better for everyone.
Community involvement is welcome and encouraged. If you want to test, suggest features, contribute code, or just explore what's possible when AI runs on your own hardware — you're in the right place.
- Governed execution — Every workflow declares its intent before running. Policy gates block bad output from propagating downstream. No silent failures — every blocked action shows exactly which rule triggered.
- Bring-your-own-AI freedom — Bring your own model, cloud provider, local runtime, OpenAI-compatible endpoint, or swarm. More AI is designed to stay useful even as the model world keeps changing.
- Local-first, cloud-optional — Run fully offline with Ollama, llama.cpp, or BitNet. Connect cloud APIs (OpenAI, Anthropic, Google, DeepSeek, and more) when you want them. Mix both in the same swarm. The Python runtime, pip packages, FFmpeg, and BitNet engine are all bundled — no internet required to get started.
- Auto-server management — Selecting any local model (Ollama, llama.cpp, BitNet) anywhere in the app automatically boots its server in the background. No manual
ollama serveor port juggling. - Multi-model orchestration — Send one prompt to 2–20 models simultaneously. Chain them, debate them, run them in a swarm, or pipeline them with a verifier scoring 0–100.
- Full provider coverage — One unified registry for every AI provider: cloud APIs, local runtimes, OpenRouter catalog (200+ models), and P2P bridge nodes.
- AI Colosseum — Calibrate your models by having them play Blackjack, Wordle, Mastermind, Battleship, UNO, Monopoly, and Chess. Capability scores, role badges, and match history included.
- Cryptographic audit trail — SHA-256 hash chain on every agentic decision. EU AI Act Art. 12/13/14/17 aligned.
 Chat — 11-model swarm: OpenAI, DeepSeek, Mistral, Google Gemini, Qwen, and local models all streaming in parallel |
 Chat — Full Swarm builder: GPU + CPU agents, cost tiers, 10 parallel API agents |
 IDE — Agent Console: 9 tool calls across 8 turns, file explorer, terminal, live tool call trace |
 Skill Builder — Design AI micro-programs in plain English or JSON DAG; validate, compile, register |
 Platform — Model Registry: Anthropic, OpenAI, Google, DeepSeek, Mistral, Ollama, llama.cpp and 339 more |
 Platform — OpenRouter Catalog: 596 models with live pricing, context windows, and benchmark scores |
 Music Studio — Local ACE-Step generation; hardware-aware model selector; lyrics + style tag sidebar |
 Picture Editor — AI image generation (ultra-realistic brindle Pitbull via Google Nano Bard); local + API models |
 AI WebView — Governed co-navigation: COE inspect/action approval gates, agent action feed, pop-out WebView2 |
 Collaborate — P2P workspace: Human Chat (encrypted) + Agent Chat side by side, WebRTC/Nostr signaling |
 Video Editor — Session recorder with export presets: YouTube HD/4K, TikTok/Shorts, Instagram Reel, LinkedIn |
 Settings — Governance Pack: Trust posture high, all 8 governance files verified (Security, Standards, Determinism, Gates, Privacy, Incident) |
 Settings — Local AI Setup: 157 model routes, 12 ready, hardware scan, intent-first model recommendations |
 Platform Hub — Connect AI, Browse Models, Bridge Agent, MCP Servers, External Apps, Usage & Cost |
 Platform — Built-in 58-chapter Manual with AI-powered Manual Expert Agent sidebar |
Models earn capability scores and role badges by playing real games. Watch AI agents compete:
| Game | Video |
|---|---|
| Blackjack | ▶ Watch Blackjack match |
| Chess | ▶ Watch Chess match |
| Monopoly | ▶ Watch Monopoly match |
| Workspace | Description |
|---|---|
| Chat | Talk to one model, compare several, or light up a full swarm across local LLMs and cloud APIs |
| IDE | Cursor/VS Code-style coding workspace with editor, file tree, terminal, preview, source control path, and governed tool agents |
| Skill Builder | Turn repeatable work into governed AI skills with typed inputs, validation, compile checks, and reusable contracts |
| Orchestra Mode | Sequential specialist pipeline: Researcher → Writer → Critic → Publisher, with visible handoffs |
| Swarm Mode | 2–20 parallel agent branches. Conductor merges outputs. Verifier scores 0–100 |
| Run Intelligence | Diff, simulate, provenance, confidence, and traceability tools for understanding what changed and why |
| Scheduler | Cron-based scheduling with budget caps, approval gates, and confidence circuit breaker |
| Unified Agent | Natural-language command bar routing intent to the right internal worker without making you hunt through menus |
| Multi-App Pipeline | Chain More AI workers with registered desktop apps so external tools can join governed workflows |
| Red Team | Deterministic security scan — prompt injection, CVE patterns, credential exposure, dependency risk |
| Folder Watchers | Inbox folder → auto-trigger pipeline → output folder. Hands-free file processing |
| Playbooks | Self-optimizing YAML workflow contracts. Compile, analyze, and auto-parallelize execution plans |
| Workspace | Description |
|---|---|
| Music Studio | Local music generation via ACE-Step (3.5B). Reference audio, LoRA styles, AI lyric sidebar |
| Voice Studio | Vocal extraction (Mel-Band RoFormer / MDX-Net). HuBERT + FAISS voice model training and conversion |
| Doc Creator | AI-driven document generation — project briefs, resumes, proposals, interview prep |
| Doc Extractor | 6-step AI pipeline: scan folder → Excel template → field map → extract → review grid → export |
| Picture Editor | Local image editing with AI-assisted tools |
| Video Editor | Screen recording, trim, crop, export with FFmpeg pipeline |
| Workspace | Description |
|---|---|
| Morning Briefing | Daily AI digest: attention items, pending workflows, changes while away, automation suggestions |
| Knowledge Base | Local RAG over PDFs, notes, and code. FAISS embeddings, chat with source citation |
| Translation | Offline translation via argostranslate. 50+ languages. Live meeting overlay mode |
| Content & Growth | YouTube scripts, SEO articles, newsletters, Twitter/X threads, LinkedIn posts, email sequences |
| Email Reputation | Local-first DMARC monitoring, domain reputation scoring, and email authentication analysis |
| P2P Discover | Discover and import pipeline optimizations from peers on your LAN |
| Game | Description |
|---|---|
| Blackjack | Card counting strategy calibration — tests rule retention and risk decisions |
| Wordle | Vocabulary and deductive reasoning calibration |
| Mastermind | Hidden-information strategy calibration |
| Battleship | Spatial reasoning and long-context stability calibration |
| UNO | Rule retention under adversarial conditions |
| Monopoly | Resource planning and negotiation calibration |
| Chess | Deep strategic reasoning — minimax vs. model |
Models earn capability scores and role badges (conductor, reviewer, adversary, planner) through Colosseum matches. Badges unlock role-specific model dropdowns throughout the app.
| Workspace | Description |
|---|---|
| AI Models | Unified model registry for Ollama, llama.cpp, BitNet, Hugging Face, LM Studio, OpenRouter, cloud APIs, and custom endpoints |
| OpenRouter | Full catalog browser (200+ models). Search, filter, one-click add. Masked key management |
| Bridge Agent | P2P compute sharing via HTTPS with TLS fingerprint pinning |
| MCP | External MCP server registry. Tool explorer. Local MCP server to expose models to other tools |
| External Apps | Register desktop apps as governed workers with declared capability contracts |
| Metrics | Real-time CPU/RAM/GPU charts. Bridge peer status. AI intelligence feed. Telemetry event log |
| Settings | Governance policies, execution profiles, API keys, developer context, hardware panels, audit chain |
| Manual | 58-chapter built-in user guide in all 11 supported languages |
Cloud APIs: OpenAI · Anthropic (Claude) · Google (Gemini) · xAI (Grok) · DeepSeek · Groq · Together AI · Mistral · Perplexity · Azure OpenAI · OpenRouter (200+ models)
Local Runtimes: Ollama · LM Studio · vLLM · llama.cpp · BitNet · Custom OpenAI-Compatible endpoints
Network: Bridge Agent (P2P / LAN / VPN) · Headless Agent · P2P Friend Share
- Internationalization — Full UI and Manual in 11 languages: English, Spanish, French, German, Chinese, Portuguese, Japanese, Korean, Arabic, Hindi, Turkish
- AI Spatial Awareness — Every agent system prompt includes current tab identity, available controls, and a full workspace navigation map so agents always know where they are
- Governance Engine — Nonce gate, local policy folder, data boundary enforcement, per-model trust profiles, policy compliance scoring. Fail-closed by design.
- Auto-Server Management — Selecting any local model automatically boots Ollama, llama.cpp, or BitNet in the background
- IDE Module System — Agents write JS modules loaded via Blob URL. Appear as live tabs with no rebuild or restart required
- Code Brain — Live repo index (tree-sitter), auto compile/test loop, invariant scanner, patch gate, file reservations, loop detector, code sprint orchestration
- Agent Performance Layer — Intent classifier, dynamic model router, semantic cache, context bus, execution traces, knowledge graph, workflow lessons, predictive prewarmer, risk predictor, adaptive swarm optimizer
- Reasoning Memory Layer — System-level operational pattern storage feeding the router and pre-execution simulation
- Pre-Execution Simulation — Dry-run before every multi-step operation: VRAM check, policy check, cost/time estimate, approval gate
- Cryptographic Audit Chain — SHA-256 hash chain of every agentic decision. Full provenance record per run
- DugBee Voice Assistant — Always-listening assistant with wake phrase detection, TTS response, workspace navigation, weather, math, distance, and custom skill integration
- Fail-closed governance — any missing governance check hard-blocks execution
- API key isolation — keys encrypted with AES-256-GCM in a dedicated SQLite database (
keys.db), separate from all other data; keys never touch the React frontend - Keys zeroized in memory after use — never stored in
Arc,Mutex, or long-lived state - TLS enforced on all outbound HTTPS — no
danger_accept_invalid_certsin production - SSRF, path traversal, and command injection prevention in all Rust handlers
- No telemetry — no phone-home, no usage tracking. Your operations are private.
- CSP:
script-src 'self' blob: 'unsafe-eval'—unsafe-evalrequired for Monaco Editor workers;blob:required for IDE module loader
| Layer | Technology |
|---|---|
| Desktop shell | Tauri 2 (Rust + WebView2) |
| Frontend | React 19 · TypeScript · Vite |
| Editor | Monaco Editor (same engine that powers many modern code editors) |
| Backend | Rust (async, Tokio) |
| Databases | SQLite via SQLx with versioned migrations (5 isolated databases) |
| Key storage | AES-256-GCM encrypted SQLite (Windows) · OS Keychain / Secret Service + AES-256-GCM fallback (macOS / Linux) |
| Local AI | Python 3.11 embeddable · torch CPU · TTS (XTTSv2, Chatterbox) · ACE-Step · faster-whisper |
| Testing | Playwright (194 tests, 0 failures) |
⬇ Download the Latest Windows Installer
No Python. No Rust. No prerequisites. Download, run, and go.
The installer bundles everything: Python 3.11 runtime, all server packages, FFmpeg, and the BitNet engine. AI model weights download on first use (hardware-specific and too large to bundle). If you have an NVIDIA GPU, a CUDA-optimized PyTorch build is swapped in automatically during onboarding.
git clone https://github.com/DougTrier/MoreAI.git
cd MoreAI
npm install
.\scripts\build-release.ps1See INSTALL.md for full build and Bridge Agent instructions. See QUICK_START.md for first-run setup in under 5 minutes.
More AI is a free, open-source project — built by one person, meant for everyone. The goal is simple: give people access to useful AI without making them hunt down runtimes, configure servers, give up local control, or pay for a dozen subscriptions.
Whether you want to test, report a bug, suggest an idea, contribute code, improve docs, or help shape model/runtime recommendations — you're welcome here. Every issue opened and every real-world test makes the app better for the next person who downloads it.
- Found a bug? Open an issue — every report is read and taken seriously
- Have an idea? Start a discussion — all suggestions are read
- Want to contribute? PRs are welcome — open an issue first so we can align on the approach
- Testing on different hardware? Share what works, what fails, and which local models perform well
- Security issue? See SECURITY.md — please don't post vulnerabilities in public issues
More AI is free and open source under the MIT License. If you want to help the project grow, you can:
- Star the repository
- Report bugs and request features
- Improve documentation
- Test new releases on different hardware
- Contribute code
- Sponsor or donate to help cover testing, infrastructure, and continued development
Support links:
- Buy Me a Coffee: buymeacoffee.com/dougtrier
- GitHub Sponsors: github.com/sponsors/DougTrier
MIT License — Copyright © 2026 Doug Trier / Trier OS.
Free to use, modify, distribute, and build on. See LICENSE for the full text.
Note: The "More AI by Trier OS" name and logo are trademarks of Doug Trier / Trier OS. The MIT license covers the code — please don't ship a fork under the same name or branding. If you build something great with it, consider contributing back so the whole community benefits.