Question 1

What is Cognisoc?

Accepted Answer

Cognisoc is an open-source project building runtimes and tooling for local, on-device LLM inference. Instead of one product, it maintains six purpose-built tools that together cover the full inference stack — a polyglot server, mobile and .NET runtimes, a composable Rust engine, an educational implementation, and a bare-metal unikernel. Everything is built on the GGUF / llama.cpp ecosystem so models are portable across the whole stack.

Question 2

What products does Cognisoc make?

Accepted Answer

Six: mullama (a local LLM runtime in Rust with native bindings for six languages and drop-in Ollama compatibility), llamafu (a Flutter plugin for on-device GGUF inference on Android and iOS), unillm (a modular type-safe Rust runtime across 47 architectures), llmdot (a .NET-native, managed-by-default local runtime), zigllm (an educational Zig codebase that teaches transformers layer by layer), and cllm (a Multiboot C unikernel that serves an llama.cpp-compatible HTTP API from bare metal).

Question 3

Which Cognisoc tool should I use?

Accepted Answer

Use mullama to replace Ollama while keeping native bindings across Rust, Python, Node.js, Go, PHP, and C/C++. Use llamafu to run a model entirely on a phone. Use llmdot for local inference inside a C#/.NET app with no native setup. Use unillm to build on a composable, type-safe Rust inference engine. Use zigllm to actually understand how transformers work by reading and writing the code. Use cllm to serve inference from bare metal with no host operating system.

Question 4

Are Cognisoc tools open source?

Accepted Answer

Yes. Every project is developed in the open on GitHub at github.com/cognisoc under permissive licenses — most are MIT, unillm is Apache-2.0, and cllm is GPL. There is no license fee.

Question 5

What is a good alternative to Ollama?

Accepted Answer

mullama is drop-in CLI-compatible with Ollama and adds first-party native bindings for Rust, Python, Node.js, Go, PHP, and C/C++, along with OpenAI- and Anthropic-compatible HTTP APIs. It is built on llama.cpp and runs GGUF models from Hugging Face, so you can embed a model in-process without running a separate daemon.

Question 6

Can I run an LLM on a phone?

Accepted Answer

Yes. llamafu is a Flutter FFI plugin that runs GGUF models entirely on-device via llama.cpp on Android (API 21+) and iOS (12.0+). It supports streaming text, chat, embeddings, vision/multimodal models, tool calling, schema-constrained JSON, and LoRA adapters — all without a server or internet connection.

Question 7

Do these tools work without a cloud API?

Accepted Answer

Yes — that is the whole point. Every runtime performs inference locally on your own hardware. Data never leaves your infrastructure, there is no per-token billing, and there is no network round-trip. (cllm is still pre-inference: its HTTP API and unikernel are shipped, but the inference engine integration is on the public roadmap.)

Question 8

Do I need to convert my models to a special format?

Accepted Answer

The tools standardize on GGUF, the quantized model format used by llama.cpp. GGUF models are widely available on Hugging Face, and because every Cognisoc runtime speaks GGUF, the same model file is portable across mullama, llamafu, unillm, and llmdot.