GizStack · Apps, data and AI
One native toolchain for apps, data, compute and AI.
Gizlang combines a language with TypeScript syntax, transactional and analytical SQL, a live web framework, replicated storage, GPU tensors and generative AI. Use the parts together in native programs, or build and host apps on GizStack Cloud.
A whole stack you can actually use
A live web app, a SQL database and an agent with tools can run in the same program. Start with one process and an embedded database; connect external services as your application needs them.
Apps that change while they run
A typed record drives stored data, validated forms and live views. Save a change with
giz live; the running program keeps named state, and open pages update in place.Transactions and analytics together
GizDB combines SQL transactions, analytical queries, full-text and vector search, key-value data and change feeds. Embed it in your app or connect through PostgreSQL and Redis protocols.
Calculate with CPU/GPU tensors, take gradients, train models, run an LLM or generate images and video. The same libraries expose ready-to-use model APIs and lower-level kernels.
Give agents typed tools, instructions, skills and memory. GizDesk brings conversations, approvals, delegation and results into an open-source workspace. Use local models or connect a model API.
The stack also includes file, object and block storage, hardware APIs, classical ML, agents with typed tools and a JavaScript/TypeScript/Python VM. Each part has its own task guide in the documentation map.
Try it here, then make it yours
The Playground runs Gizlang in your browser. To see stored data and live pages working together, run the todo app locally:
npm install -g @gizai/cli
giz init my-app --template todo
cd my-app
giz live main.gizOpen http://127.0.0.1:8080 in two tabs. Add a todo or check it off: both tabs update from the same stored record.
Edit a source file while the app runs. The template includes the model, page and server source.
The CLI needs Node.js 20.12+. The native toolchain runs on Linux x86-64 (glibc 2.35+), including WSL 2. The first run downloads about 190 MB; macOS support is planned. Installation details →
Three names, three jobs
| Product | What it is | Where to start |
|---|---|---|
| Gizlang | The open-source language and toolchain, with GizWeb, GizDB and AI libraries. Build your own apps and agents. | Language tour |
| GizDesk | The open-source AI agent platform and workspace. Connect tools, models and knowledge; work with agents and their results. | Run GizDesk |
| GizStack | The cloud service built on this stack, with hosted apps, cloud resources and managed services. The service source is private. | Explore the cloud |
Gizlang and GizDesk are MIT-licensed and run without a GizStack account. Cloud hosting is in preview; the roadmap lists the state of each cloud capability. Work to bring GizDesk and the cloud workspace onto the same agent platform is in progress.
GizDesk: more than a chat window
Use an agent with MCP tools, skills, memory, reference material and scheduled tasks. Review tool calls, compare models, and keep documents or code beside the conversation. Voice and character video use configured media models.
giz init my-desk --template desk
cd my-desk
giz run main.gizModel downloads and hardware requirements depend on the model. You can connect an OpenAI-compatible API instead. External tools and media workers need configuration. The GizDesk guide describes setup, authentication and current limits.
Why build the layers together?
Less application glue. The built-in data flow connects typed records, queries, validated forms and live views. You can build it without a separate ORM, JSON API and client state store.
Native programs. Gizlang compiles .giz source through LLVM. The web server, embedded database and AI features
are libraries of the same toolchain. The VM for JavaScript, TypeScript and Python is a separate execution path;
compatibility with those ecosystems is still developing.
A shorter edit loop. giz runs, checks, tests, formats, builds and live-reloads. The compiler checks types and
errors before changed code runs. See the developer tools →
Measured, not promised
Comments app against Next.js: Thread page, one CPU (Next.js: one process). Next.js 16: 197 req/s.
Comments app against Next.js: JavaScript sent to the browser. Next.js 16: 592,598 B (196,717 B gzip).
LangArena: Runtime score. Rust: 90.6. C: 90.2.
These are our published measurements of particular workloads, not official leaderboard results or a promise about every application. Each benchmark table gives its date, machine and method, including slower workloads and reproduction limits. Some measurements predate their suites and have not been recorded again by them; some suites require the private cloud sources.
Build in the cloud or on your computer
LocalInstall the toolchain
Build a live app with typed data, forms and an embedded database. The source is yours to read and change.
npm install -g @gizai/cli
giz init my-app --template todo && cd my-app
giz live main.gizNode.js 20.12+. Native toolchain: Linux x86-64, glibc 2.35+ (including WSL 2). First download: about 190 MB. macOS support is planned.
Explore the stack
- GizlangAvailable
A compiled language with TypeScript syntax, native code and live code swap.
- GizDBAvailable
One engine for transactions, analytics, key-value data and change feeds. Embedded, or a PostgreSQL and Redis server.
- GizWebAvailable
Server-rendered apps from one typed model, with a browser kernel of about 10 KB and live updates that keep state.
- StoragePreview
Objects, files and block volumes in one layer, with the S3 API and a POSIX mount.
Cloud services and their current status
- Giz AppsAvailable
Describe an app or start from a template, edit its code, deploy it live in about two seconds.
- Giz ComputerComing
Persistent computers and volumes, in our regions or on your servers.
- Giz DrivePreview
Files and documents with versions, sharing, search and sync.
- Giz AgentPreview
AI agents that do the work inside your rules, approvals and budgets.
- Giz OntologyComing
Business objects, relations and Actions, with ready industry packages.
- Giz DataComing
Analytics, dashboards and BI, directly on GizDB.
- AI GatewayPreview
One metered endpoint for AI models, with budgets and caps.
GizStack combines app hosting, AI and cloud resources. Each service above links to its capabilities and availability. How the platform fits together →
Read the code. Check the claims. Try a small app.
Browse the MIT-licensed language, libraries and GizDesk. Build from source or start from a template.
Compare concrete workloads, inspect the test conditions and see where more work remains.
Documentation for you and your agent
Every docs page is available as Markdown. Compiler checks, outlines and capabilities also have structured output.
Start with one working program