Agul is a terminal coding agent written in Rust. Run it inside a project, talk to a model, and let it read, edit, execute, and verify work without leaving the terminal.
- Full-screen TUI with visible reasoning, tool activity, steering,
/stop, session history, and an editable multiline input. - Four built-in tools:
read,write,edit, andshell. - DeepSeek, GLM Coding Plan, ChatGPT/Codex account login, and local OpenAI-compatible endpoints.
- Persistent sessions,
--continue/--resume, per-response token usage, KV reporting, and a versioned Usage Ledger. - Project and user Skills, prepared Plugins, and ARI for optional coordination.
Web Search and multi-agent coordination are optional AgentKube extensions; they are not hidden built-ins in the minimal runtime.
Agul itself does not require Bun, Node.js, or Agulater. 0.6.0-rc.1 is a
prerelease, so its installer URL is deliberately pinned instead of using the
stable-only GitHub latest alias:
Linux or macOS:
curl -fsSL https://github.com/storious/agul/releases/download/v0.6.0-rc.1/install.sh | shWindows PowerShell:
irm https://github.com/storious/agul/releases/download/v0.6.0-rc.1/install.ps1 | iexThe same release page contains archives for Windows x64, Linux x64, macOS x64, and macOS ARM64.
Agulater is optional. Use it when you want managed runtime updates, Catalog packages, Skills, or Plugins. Its published installer contains a standalone executable, so normal users do not need Bun, Node.js, or npm. Bun is only a build and source-development tool for Agulater.
Contributors can build the same runtime directly:
git clone https://github.com/storious/agul.git
cd agul
cargo build --release --lockedRun target/release/agul on Linux or macOS, or
target\release\agul.exe on Windows.
Agul defaults to DeepSeek. Set the key, enter a project, and run agul; the
workspace defaults to the current directory.
export DEEPSEEK_API_KEY='<key>'
cd my-project
agulPowerShell uses $env:DEEPSEEK_API_KEY = '<key>' and the same agul command.
Other connections are explicit:
| Connection | Command |
|---|---|
| GLM Coding Plan | agul chat --provider glm --reasoning-effort high |
| ChatGPT/Codex quota | agul account login, then agul chat --engine codex |
| Local/OpenAI-compatible | agul chat --base-url <url>/v1 --model <model> |
For one non-interactive turn:
agul chat --prompt "find the failing test, fix it, and rerun it" --jsonUseful interactive commands are /status, /skills, /usage, /compact,
/sessions, /stop, and /exit. Resume the latest workspace session with
agul chat --continue or choose one with agul chat --resume.
- Agulater installs and prepares optional
.agentspackages. - AgentKube supplies optional Skills, Plugins, specialist agents, Web Search, and multi-agent coordination.
- ARI lets those components drive the same Agul runtime instead of introducing another agent CLI.
Basic Agul use needs none of them.
Apache-2.0. See LICENSE. Runtime dependency licenses are summarized in THIRD_PARTY_NOTICES.md.