Aizen — the terminal-native coding agent that lives on your machine

One static binary. No Node, no Python, no Docker, no cloud account. Download aizen, point it at any OpenAI-compatible endpoint, and you have a full agentic coding partner in your terminal — one that reads and edits your code, runs your shell, verifies its own work, remembers how you like things done, and keeps working when you leave (drive it from Telegram on your phone).

The command is aizen. This repository is the download channel — prebuilt binaries and the install scripts. (The source lives in a separate private repo; this repo is generated.)

Demo

liveaizen.mp4

Why aizen

• Zero-friction install. A single self-contained executable — no runtime, no containers, no npm i -g dependency tree. Grab the binary, run it. Pure-Rust with rustls-only TLS, so there's no OpenSSL or native toolchain to fight.
• Bring your own model. Works with any OpenAI-style /chat/completions API — OpenAI, OpenRouter, a local llama.cpp/vLLM server, or an Anthropic-backed gateway. You pick the provider; you're never locked to one lab. Context window auto-detects; cost is tracked live.
• It actually finishes tasks. A real tool-using loop reads/edits files, runs shell, searches the web, and — crucially — verifies before it claims done (runs your typecheck/tests and fixes what broke). Parallel reads, serial + approval-gated writes.
• It remembers you. A self-learning memory brain, an evolving persona, a durable operating SOUL, and reusable skills mean the agent gets more useful over sessions instead of resetting every time.
• It runs where you aren't. aizen serve turns any Telegram (or Discord) bot into a remote control for the agent on your machine — approve risky edits from your phone with a tap. Host it 24/7 as a systemd service on a VPS.
• Safe by construction. File/shell tools are confined to the working dir; a hard safety floor blocks catastrophic commands even under auto-approve; secrets are written owner-only and never printed. You stay in control of every destructive step.

Install

One line — grabs the latest optimized binary for your OS and puts aizen on your PATH:

# Windows (PowerShell)
irm https://raw.githubusercontent.com/dawnofcd/aizen/main/install.ps1 | iex

# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/dawnofcd/aizen/main/install.sh | sh

Then open a new terminal and run:

aizen config     # base URL → API key → pick a model
aizen            # launch the REPL — just type

That's it — after aizen config, every command works with zero env vars. Override the install directory with $env:AIZEN_INSTALL (Windows) or $AIZEN_INSTALL (Unix). The Windows .exe is unsigned — if SmartScreen warns, choose More info → Run anyway.

Or download a binary by hand

Grab the asset for your platform from the latest release:

Platform	Asset
Windows x86-64	aizen-<ver>-windows-x86_64.exe
Linux x86-64	aizen-<ver>-linux-x86_64
macOS Apple Silicon	aizen-<ver>-macos-aarch64

On Linux/macOS make it executable: chmod +x aizen-* && ./aizen-….

Feature map

Area	What you get
Unified REPL	One chat+agent loop (no mode switch), a live status HUD (model · tokens · turn · % context bar), a real line editor with history, image/vision input, and a retained full-frame TUI with responsive Markdown, tables, and diagrams.
Agent loop	Read/edit/glob/search/shell tools · parallel reads · approval-gated writes · a verify gate (auto typecheck/test + one fix turn) · clarify-don't-guess · sub-agent dispatch (task) · LSP-powered symbolic edits.
Memory brain	Offline, BM25-ranked, Unicode-aware (Vietnamese-safe) retrieval that evolves from reuse — no LLM, zero extra tokens. #text remembers a fact in one keystroke.
Persona + SOUL + skills	A swappable persona with evolving self-memory (Generative-Agents style), a durable SOUL identity above every persona/project, and skills the agent loads on demand and learns automatically after real work.
Multi-agent	aizen workflow fans out role-scoped sub-agents (mixture-of-agents) and merges the results; /workflows shows a live status registry of running tasks and workflows.
Remote control	aizen serve (Telegram) / aizen discord serve — full / command menu, multi-bot hosting from one daemon, per-chat context, phone approvals, systemd self-host. Plus OS-scheduled aizen cron runs.
Extensibility	MCP servers (stdio/HTTP, OAuth 2.1 sign-in for Linear/Notion/Slack/Gmail/Atlassian) · custom markdown slash-command macros · outbound notify channels.
Web + browser	web_search / web_fetch / web_crawl (katana-style crawler, SSRF-guarded) and opt-in CDP browser tools that drive a real Chrome/Edge — all pure-Rust, no headless engine bundled.
Safety + recovery	Workspace confinement · hard command floor (survives /yolo) · owner-only secret files · crash-recoverable Git checkpoints (/timeline · /checkpoint) · per-turn MCP schema pinning · per-conversation browser isolation.

Quick start

aizen config                     # interactive: endpoint, API key, model
aizen                            # REPL — just type
aizen agent "add a --version flag and update the help"   # one-shot task
aizen serve                      # run the Telegram/Discord bot surface

Configuration lives under your user config dir (~/.aizen/); API keys never leave your machine (they go only to the endpoint you configure).

License

MIT — see LICENSE.