aivyx-coder · ManualContents ↓
Welcome to aivyx-coder
aivyx-coder is a coding agent that lives in your terminal and works with local models only — Ollama, Lemonade Server, Docker Model Runner, llama.cpp, vLLM or Jan. You describe a change; the model reads your project, edits files, runs your tests and uses git, and you approve each step. Nothing goes to a cloud API.
What makes it different
- Local, always. It talks only to model servers on your own machine or network. There is no cloud mode to switch on by accident.
- Built for a model that may be wrong. Every file edit and command
waits for your approval, with the exact diff or command shown. Your
secrets (SSH keys, cloud credentials,
.envfiles) are off-limits, and on Linux every command runs inside a kernel sandbox. - Easy to take back. Each turn is snapshotted in git, so
/undorewinds everything the assistant just did — edits and command effects together. - Made for small models too. A repository map, a choice of edit formats, and careful context management help 4–30B models do real work.
How to read this guide
Start with Install and first run and A working session; together they cover everyday use. Then pick what you need:
- plan before acting — Plan mode
- review, undo, commit and test — Undo, diff, commit and test
- come back to a conversation — Sessions
- several models — Models and routing, Local model servers
- let it work alone — Autonomous and advanced modes
- use it from Zed or VS Code — Editor integration
Every command, setting and tool is in the reference.