ZYVO — OpenCode (Ready-to-Use AI Coding CLI)

ZYVO is a fully customized AI coding agent — install and it just works. Native Android build for Termux (no proot, no glibc), with automatic install on Ubuntu proot / Debian / Chromebook / WSL / macOS as well.

Install

curl -fsSL https://raw.githubusercontent.com/zyvo9/zyvo/main/install.sh | bash

Or from PyPI (same installer, layer fetched through pip):

pip install zyvo && zyvo

That's it — one command, zero config. No pkg install needed beforehand: python, ripgrep, unzip and git are auto-installed if missing, and skipped if already present. The installer:

• shows a single live progress line ([████░░░░] 42% | 18.4/29.1 MB | 1.1 MB/s) for the core download and for every long step, so you always know where it is
• auto-installs missing dependencies (pkg/apt) and skips what you already have
• keeps a built-in free AI key (zero config — override with ZEN_API_KEY)
• is safe to re-run — updates the core delta-wise (0 MB if nothing changed)
• resumes interrupted downloads and verifies the core with SHA256

Update

zyvo update

Termux: Use the F-Droid version (the Play Store build will NOT work). Other platforms: Ubuntu proot (Andronix / UserLAnd), Debian, WSL, macOS — same command.

Device support

Full-power mode

The AI works at full strength:

• DeepSeek model by default (deepseek-v4-flash) — stable + fast. Other tiers: oc-settings model lightning|mid (nemotron-ultra sometimes has provider errors — experimental)
• Full-power agent — the AI is instructed to: understand first → plan → execute step by step → verify by itself → keep going until the WHOLE task is done. No placeholders or TODOs are delivered; web search brings in current information
• All tools open — no permissions, web search/fetch enabled — nothing holds the AI back

Permissions

None. ZYVO shows no permission prompts — the AI runs commands, edits files, and does web work on its own (beginner-friendly design).

Overview

Usage

zyvo                 # start the AI — ready to work right away
zyvo session         # list sessions
zyvo session proj1   # new/resume session "proj1" — stored in zyvo/proj1/
zyvo update          # delta update check (0 MB if core is unchanged)
zyvo preview         # local server + auto-open any page in the browser
zyvo uninstall       # cleanly remove ZYVO (keeps your projects)
oc-settings          # model tier menu

Sessions (separate projects)

Each session has its own folder and chat history:

/storage/emulated/0/zyvo/
├── seson1/     ← zyvo session seson1  (files + chats live here)
└── seson2/     ← zyvo session seson2

zyvo session seson1 resumes that session's conversation — the AI remembers what was said before (--continue). Names may only contain a-z 0-9 . _ -.

Restart zyvo after changing the model.

Reinstall / Update (delta)

zyvo update — or just run the same install command again.

Delta update system: the core binary carries a version stamp. If the release is unchanged, the core is not downloaded again (0 MB) — only the ZYVO layer (config / skills / commands, a few KB) is refreshed. If the core changed, a full download happens.

User settings are preserved: model choice, provider key — updates merge config, they don't overwrite it.

Performance (phone lag / heating)

Lag and overheating were mainly caused by working at the storage root (/storage/emulated/0) — scanning + snapshot-tracking thousands of files (DCIM, WhatsApp, Android/) burned CPU. Now:

• Dedicated workspace — default /storage/emulated/0/zyvo (auto-created). The AI works there; the whole storage is no longer scanned
• Snapshot off (snapshot: false) — no separate git repo at the storage root. You can re-enable it in config if you want undo
• Watcher ignore — DCIM / Pictures / WhatsApp / Android... are not watched
• Autoupdate off + sharing disabled — no background network/CPU work

Extra tips: press Ctrl+K to disable TUI animations (great for battery), work in a specific folder with zyvo /storage/emulated/0/Documents.

Troubleshooting

• zyvo: command not found → pkg install ripgrep + restart Termux
• Model error → source ~/.bashrc (check key: echo $OPENCODE_API_KEY)
• Install failure → check internet and rerun bash install.sh (downloads resume)
• 32-bit phone → no binary available; use a 64-bit device or proot

Full details (customization, keys, device support) — the installer, config, and scripts are all open source in this repo.