An AI-native full-stack programming ecosystem built around the Jac language, enabling frontend, backend, AI integration, and cloud deployment from a single file.
Jaseci is an AI-native full-stack development ecosystem built around the custom Jac programming language, led by Professor Jason Mars at the University of Michigan with NVIDIA Inception membership. Its core design goal is to eliminate tech stack fragmentation in modern AI application development — developers no longer need to switch between Python backends, JavaScript frontends, YAML deployment configs, and prompt engineering, but can accomplish everything in a single .jac file.
Language and Compilation
Jac features Python-style syntax, with a compiler that translates .jac source code into Python bytecode, JavaScript, and native machine code (C-ABI compatible), directly bridging the full ecosystems of PyPI, npm, and C libraries. The project claims an average 10x reduction in code volume compared to traditional stacks.
AI-Native Programming Model
Meaning Typed Programming, implemented through the byllm plugin, is Jaseci's core differentiator. Developers use the def func() -> ReturnType by llm() syntax to define functions, and the compiler automatically extracts semantic information from function names, parameter types, return types, and docstrings to construct prompts — calling LLMs without writing any prompts manually. Combined with built-in graph nodes (node) and traversal agents (walker), it is naturally suited for building multi-step Agentic AI systems.
Full-Stack and Deployment Integration
The jac-client plugin supports writing backend logic and React-style frontend components in the same file, with frontends directly calling backend functions, eliminating REST/HTTP/CORS boilerplate. The jac-scale plugin enables zero-config deployment from local to production: jac start app.jac --scale deploys to Kubernetes with one command, automatically configuring Redis, MongoDB, JWT auth, Swagger docs, load balancing, and health checks — no Dockerfile or K8s manifest needed.
Graph-Native Data Modeling
Jaseci features a built-in graph engine as its core programming paradigm. node defines graph nodes, walker defines agents that traverse and execute on the graph, and graph structures are manipulated via ++> and --> operators with automatic persistence (SQLite for development, MongoDB for production).
Current version is jaseci 2.3.13, including jaclang 0.14.0, byllm 0.6.4, jac-client 0.3.12, jac-scale 0.2.14, and other sub-packages, all independently installable from PyPI. Licensed under MIT.