What it is
CodeZaiku points a language model at real systems. It edits code, fixes a broken service stack, reviews a diff, reports on security findings, and answers questions from the web. It talks to any OpenAI-compatible server: a model on your own machine (llama.cpp, Ollama, LM Studio, vLLM) or a hosted API.
It is one command:
codezaiku chat ./repo # talk to it: it asks before changing anything, /undo steps back
codezaiku fix my-stack # diagnose a broken compose project and repair it
codezaiku code ./repo "make the tests pass"
codezaiku review ./repo # read-only: findings, no edits
codezaiku secure my-host # posture and intrusion findings
codezaiku research "question" fan # parallel researchers, a critic, an answer with sources
codezaiku watch my-stack # keep looking, fix when something breaks
By default it proposes changes rather than making them.
In chat, /research sends a question to the researchers, and a research memory keeps
what they found between runs. codezaiku install researchzosho installs the ResearchZosho research
library and connects the chat to it.
Another program can drive it three ways: an MCP server, an ACP agent, or a subprocess that takes a task and writes JSON to stdout.
Install
Current release: 0.3.12 — every artifact, with checksums, is on the download page.
Linux and macOS:
curl -fsSL https://codezaiku.org/install | sh
Windows, from PowerShell:
irm https://codezaiku.org/install.ps1 | iex
Both scripts download the release from GitHub and check it against that
release’s SHA256SUMS before installing. If the hash does not match, they stop.
Only the script comes from this domain — the binary and the checksums both come from the
GitHub release. If you would rather not pipe a script into a shell, download it and read it first;
it is short.
You need an OpenAI-compatible model server; CodeZaiku ships no model weights. You also
need Java 21 or newer, unless you take a build that carries its own Java runtime: the install scripts
pick that build when Java is missing. On Windows you also need Git for Windows, because it shells out
through bash. Run codezaiku doctor and it will tell you what is missing and how to fix it.
Which model, by VRAM (the graphics card's memory, not the computer's RAM), is on the models page;
codezaiku model serve install sets that one up on this machine, on demand.
Documentation
- Architecture — how it works.
- Platforms — Linux, macOS, Windows and WSL2.
- Limitations — the measured numbers.
- Driving it from another agent — the contract and what it returns.