Orbit
A local, CPU-only agentic AI runtime for Linux that combines chat, model-selected tool use, file workflows, and static artifact analysis in one terminal client, built on a vendored llama.cpp backend.
Orbit is a Python-based local AI runtime built for CPU-only machines, qualified on Linux x86_64, that combines chat, tool use, file workflows, and static artifact analysis in a single terminal client. It manages tools, sessions, evidence, verification, and bounded repair itself, while an integrated, vendored llama.cpp backend handles tokenization, inference, streaming, and model state, so no separate llama.cpp installation or external inference server is required.
It ships a curated registry of verified model and quantization combinations, currently including Gemma, Ornith, Qwen 3.6, Qwen3-Coder, and Qwen3.8 variants, and the README distinguishes between a model being qualified to chat and call tools correctly and being qualified for its ANALYSIS explanations to be semantically correct, noting that even its most-tested analysis profile still has retained reports containing factual errors.
Building from source requires Python 3.11 or newer, Git, CMake, a C/C++ toolchain, and, for the ANALYSIS feature, bubblewrap; a local orbit server process handles inference and a separate orbit terminal client connects to it.