ZYVA

AI-powered Cloud IDE — build and ship apps from your browser.

Try Cloud IDE → · Download Desktop · Docs · zyva.dev

---

---

Download (Desktop App)

Latest release: v0.4.0 — all releases →

Platform	Download
🪟 Windows	ZYVA.Setup.0.4.0.exe
🍎 macOS (Apple Silicon)	ZYVA-0.4.0-arm64.dmg
🐧 Linux (.deb)	zyva-desktop_0.4.0_amd64.deb
🐧 Linux (AppImage)	ZYVA-0.4.0.AppImage

Builds are currently unsigned — your OS may warn on first launch. Prefer the browser? Just use the Cloud IDE (no install).

Self-hosting? A container image is published to GitHub Packages (GHCR): docker pull ghcr.io/titanxlayer/zyva:latest

---

What is ZYVA?

ZYVA is a Lovable-style AI coding environment with two modes:

	Cloud IDE	Desktop App
Access	Browser — no install	Electron — local
Execution	WebContainer (browser) + E2B sandbox	Local machine directly
Isolation	0G TEE + Firecracker VM per session	User's own machine
Storage	0G persistent storage	Local filesystem
AI	0G Private Computer (TEE-attested)	0G Private Computer

Cloud IDE is the primary product. The desktop app stays available for teams that need fully local control.

---

AI Inference — 0G Private Computer

All inference runs on 0G Private Computer — OpenAI-compatible API, every request inside a TEE. No BYOK required.

Model	Context	Best for
minimax-m3 ⭐	1M	Multimodal, default
glm-5.1	207K	Long-horizon coding
qwen3.7-max	1M	Function calling
qwen3.6-plus	1M	Multilingual
deepseek-v4-pro	1M	Agentic coding

---

Stack

Layer	Technology
UI shell	Next.js 16 (App Router), React 19, Monaco editor, Tailwind v4, Zustand, Framer Motion
Inference	0G Private Computer (pc.0g.ai) — OpenAI-compatible, TEE-attested
Auth	NextAuth v5 — Google, GitHub OAuth + SIWE wallet (0G Chain)
Database	PostgreSQL (Prisma v7) — users, sessions, projects, traces
Sandbox	E2B — on-demand, build/install only, torn down after task
Git	Real git commit + push to GitHub from the IDE Source Control panel
Embeddings	Qwen text-embedding-v4 (DashScope) or local Ollama
Rerank	qwen3-rerank
Vector store	Local file-backed (default) or Qdrant
Observability	Local trace store + optional Langfuse
Desktop	Electron wrapper around Next.js standalone build

---

Repository Map

zyva-app/
├── src/
│   ├── app/                    Next.js App Router
│   │   ├── api/
│   │   │   ├── agent/
│   │   │   │   ├── run/        Iterative agent loop (tools) + self-healing build (SSE)
│   │   │   │   └── stream/     Multi-agent graph (Architect → Review) (SSE)
│   │   │   ├── ai/             Chat completions (streaming, 0G PC / ZYVA)
│   │   │   ├── auth/           NextAuth route handler
│   │   │   ├── checkpoints/    List snapshots + rollback
│   │   │   ├── git/            Real git commit + push + clone from GitHub
│   │   │   ├── index/          Semantic index of a project
│   │   │   ├── sandbox/        E2B preview (cloud) + build tasks + kill
│   │   │   ├── terminal/       Secure command execution
│   │   │   ├── traces/         Observability trace list
│   │   │   └── workspace/      File tree, save, create project (+ design retrieval), browse
│   │   ├── auth/               Sign-in + error pages
│   │   ├── docs/               Public documentation pages
│   │   └── icon.png            App favicon (ZYVA logo)
│   │
│   ├── components/
│   │   ├── AgentSwarm.tsx      Right panel: Swarm + AI Chat + Agent Loop toggle
│   │   ├── ChatComponents.tsx  Markdown, action cards, TEE badge
│   │   ├── LivePreview.tsx     WebContainer preview (+ E2B "Open in browser", cloud only)
│   │   ├── MonacoCodeEditor.tsx Editor + Prettier + Emmet + snippets
│   │   ├── SidebarPanel.tsx    Explorer, Source Control, Extensions
│   │   └── TerminalConsole.tsx Secure terminal with TEE badge
│   │
│   ├── engine/                 ← Core runtime
│   │   ├── config.ts           Central env/config
│   │   ├── tools/              Agent tool schema + executor (read/write/edit/grep/run/db)
│   │   ├── orchestrator/
│   │   │   ├── runAgentLoop.ts ReAct loop: reason → tool call → observe → repeat ⭐
│   │   │   ├── graph.ts        Multi-agent graph (Architect/Frontend/Backend/Review)
│   │   │   ├── runAgent.ts     Single bounded reasoning step
│   │   │   └── agents.ts       Agent role definitions
│   │   ├── build/              Self-healing build loop (run build → fix errors → repeat)
│   │   ├── db/                 Per-project SQLite (better-sqlite3, lazy)
│   │   ├── design/             Design-template library retrieval (embeddings)
│   │   ├── repomap/            Aider-style repo map (symbol outline)
│   │   ├── retrieval/          Chunker + vector store + embed/rerank query
│   │   ├── execution/          E2B sandbox (preview + build, cloud only)
│   │   ├── git/                Real git operations (commit/push/clone)
│   │   ├── patch/              SEARCH/REPLACE + snapshot/rollback
│   │   ├── providers/
│   │   │   ├── ogpc.ts         0G Private Computer — primary inference ⭐
│   │   │   ├── zyva.ts         ZYVA (DO Inference Router) — internal/locked
│   │   │   ├── dashscope.ts    Qwen embeddings + rerank
│   │   │   ├── ollama.ts       Local embeddings (optional)
│   │   │   ├── gateway.ts      Embedding gateway (optional)
│   │   │   └── types.ts        Provider interfaces (+ tool support)
│   │   ├── security/           Command policy (allow/approve/deny)
│   │   ├── observability/      Trace store + Langfuse forwarder
│   │   └── tee/                Honest TEE attestation state
│   │
│   ├── lib/                    auth-guard, github, prisma, wallet (SIWE),
│   │                           workspace-isolation, extensions, snippets, file-icons
│   ├── auth.ts                 NextAuth v5 config (Google/GitHub/SIWE)
│   ├── middleware.ts           Edge auth middleware
│   └── store/useIdeStore.ts    Zustand global state
│
├── templates/                  Injected into every new user project
│   ├── CLAUDE.md · AGENTS.md · DESIGN.md   Guidance files
│   └── design-library/         74 curated DESIGN.md systems + embeddings (retrieval)
│
├── landing/                    Static landing page (zyva.dev)
├── desktop/                    Electron wrapper (main.js, prepackage.mjs)
├── gateway/                    Standalone embedding gateway (server mode)
├── prisma/schema.prisma        DB schema (users, sessions, projects, traces)
├── Dockerfile                  Cloud IDE container image (→ GHCR / Packages)
└── .github/workflows/
    ├── ci.yml                  Build + lint on push
    ├── release.yml             Cross-platform desktop builds on tag
    └── docker.yml              Container image → GitHub Packages (GHCR)

---

Getting Started

git clone https://github.com/titanxlayer/zyva
cd zyva-app
npm install
npx prisma generate
cp .env.example .env.local   # add your 0G PC API key
npm run dev                  # http://localhost:3000

Environment

Var	Purpose
OG_PC_API_KEY	0G Private Computer — pc.0g.ai
OG_PC_BASE_URL	https://pc.0g.ai/v1
OG_PC_MODEL	minimax-m3 (default)
E2B_API_KEY	E2B sandbox — e2b.dev
DATABASE_URL	PostgreSQL connection string
NEXTAUTH_SECRET	Random secret — openssl rand -base64 32
NEXTAUTH_URL	https://your-domain.com
GOOGLE_CLIENT_ID/SECRET	Google OAuth
GITHUB_CLIENT_ID/SECRET	GitHub OAuth
DASHSCOPE_API_KEY	Qwen embeddings + rerank
ZYVA_WORKSPACES_ROOT	Per-user workspace root directory

Build desktop

NEXT_STANDALONE=1 npm run build
cd desktop && npm run dist

---

Security Model

• LLM never executes shell directly — all commands go through the policy layer
• allow → auto-run safe commands (npm install, tsc, eslint)
• approve → requires user confirmation (rm, docker, chaining)
• deny → blocked (rm -rf, sudo, fork bombs)
• Per-user workspace isolation — paths validated server-side
• E2B sandboxes scoped per user session, torn down after task
• 0G TEE attestation recorded for every inference request

---

License

MIT — see LICENSE

Third-party components retain their licenses. See NOTICE.