Tablif TownSofuu: The general purpose Runtime for AI
For when you need to run local llm in javascript without python.
Run local language models, vector memory, and autonomous AI agents in one process for JavaScript and TypeScript developers.
What it does
Continuous QTSQ vector memory
Stores cognitive vector memory directly inside encrypted local files with exponential decay.
“In-process cognitive memory stored in AES-256-GCM encrypted .qtsq files. Ebbinghaus exponential decay, episodic consolidation, and zero external DBs.”
Native MCP engine
Connects Model Context Protocol clients and servers without Python environments.
“Built-in Model Context Protocol client and server. Connect external tools, IDEs, and local services via standard JSON-RPC without Python virtual environments.”
In-process local LLM inference
Executes GGUF and SafeTensors models directly on CPU SIMD or Metal GPU.
“Dense matrix math on Metal GPU shaders and ARM NEON / AVX2 kernels, with zero-copy mmap tensor loading for GGUF and SafeTensors.”
R7 security firewall
Enforces deterministic security policies on tools, filesystem, network, and shell actions.
“Deterministic security boundary strictly evaluating tool execution, shell access, filesystem read/write paths, network egress, and MCP invocations.”
Embeddable C ABI engine
Allows embedding the runtime inside mobile applications on iOS and Android.
“QuickJS pure interpreter with zero-JIT execution. Fully compliant with Apple iOS App Store”
What Sofuu doesn't do
- It doesn't support model training or fine-tuning.
- It doesn't provide multi-GPU distributed serving.
- It doesn't support Node N-API native addons or worker threads.
- It doesn't support Windows continuous integration.
Questions people ask
4 answers, each tied to a line on sofuu.xyz.
What is Sofuu?
Sofuu is a self-contained JavaScript and TypeScript runtime in a single binary that includes in-process local LLM inference, autonomous agents, native MCP, and encrypted vector memory.
Source: sofuu.xyzHow fast is Sofuu's cold start time?
The median cold start measured on macOS Apple Silicon is 22.5 ms including process spawn.
Source: sofuu.xyzCan Sofuu run on iOS devices?
Yes, Sofuu can be embedded in iOS applications using libsofuu because it uses a pure interpreter without JIT compilation.
Source: sofuu.xyzWhat inference provider should be used for production local LLM execution?
For production use, the llamacpp provider should be used while the built-in native inference engine remains experimental.
Source: sofuu.xyzTablif reads sofuu.xyz and writes this page; every line comes from the product's own pages. The founder can correct or hide any line. Tablif publishes one welcome page; after that, new pages go live only after the founder approves them. The Tablif Town game below never changes what is written here.
Sofuu is #90 in Tablif Town
0 residents · 0 campaigns sent. Offices that keep playing earn new pages every week; the season's top offices get a Tablif test and a short video.