zchee

Ganja Code — AI skill for Claude Code

AI community

A cloud-native AI coding agent that lives in your terminal — written in Rust.

How to install Ganja Code

This entry records only its repository, not the path inside it, so there is no exact command to give. Open zchee/ganja-code and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Ganja Code does

A cloud-native AI coding agent that lives in your terminal — written in Rust.

Alternatives in AI

Codeseek — Rust-powered code intelligence CLI for AI coding agents 764 ★
Pebrel — AI-native, GPU-accelerated terminal emulator for Windows with SSH, persistent sessions, split panes, and first 670 ★
Nebula — AI-native, GPU-accelerated terminal emulator for Windows with SSH, persistent sessions, split panes, and first 587 ★

README

ganja-code

**ganja** is a cloud-native AI coding agent that lives in your terminal: the only agent process on your machine should be the one you are talking to.

Rust, a ratatui TUI, permission-gated tools, and a team model where agents outlive the call that started them — including other vendors' CLIs (Claude Code, Codex, Grok, and agy), driven as teammates.

[!IMPORTANT] **ganja-code is under active development and not yet stable.** Features and configuration may change without notice.

Design

Everything ganja fans out — subagents, teammates, whole parallel workstreams — is headed for the cloud, with ganja as their control plane. Your terminal stays the cockpit, and three things follow from that.

Your machine stops being the ceiling. How large a fleet you run is no longer a question about your laptop's fans, its RAM, or how long its battery lasts.
Isolation is the default, not a setting. An agent sandboxed in the cloud has no reach into your credentials, your SSH agent, or the rest of your home directory. The blast radius of a poisoned prompt or a bad tool call is whatever you sent it, rather than everything the process could touch — a distinction most of this field still treats as an afterthought.
Scale you cannot buy locally. 10,000 subagents at once is a budget question rather than a hardware one.

[!NOTE] None of the cloud behaviour above is implemented yet; it is the design this is being built toward. Teammates currently run on your own machine, in-process or in a tmux pane beside you, with the reach that implies.

If it looks useful, a **Star** is welcome. Releases are not published yet, so **Watch** the repo to follow the work as it lands.