Terminal
Work in the fullscreen TUI, or run a single prompt with arabica -p in a script.
Open source · Apache-2.0 · Rust 1.88
Arabica is a headless AI coding agent built around one command and event protocol. Use it in your terminal, connect an editor through ACP, or bring your own client to its HTTP host.
Clients submit typed commands. The session layer validates and sequences each run. The runtime orchestrates model calls and tools, then emits canonical events. Every host shares that same contract.
Read the documentationThe runtime stays headless. You decide where it meets your work.
Work in the fullscreen TUI, or run a single prompt with arabica -p in a script.
Connect an ACP client, including Zed, through arabica acp over stdio.
Submit commands and subscribe to events through an Axum HTTP and SSE server.
API overviewInside the roast
Protocol types, session sequencing, provider adapters, and tool execution each have their own crate. Unsafe Rust is forbidden across the workspace.
Explore the architectureChoose a verified release download or build the published Rust package. Both install the arabica command.
macOS 14+
Download the native macOS app for your Mac, unzip it, then move Arabica.app to Applications.
Apple silicon (M1–M-series) ↗
Intel Mac ↗
The app includes the Arabica agent. Configure a provider in Settings on first launch. The current v0.4.0 package is unsigned and not notarized; macOS may block it.
macOS + Linux
The installer detects your CPU, checks the release SHA-256 file, and places the binary in ~/.local/bin.
curl -fsSL https://haowei2000.github.io/Arabica/install.sh \
-o install-arabica.sh
sh install-arabica.sh
~/.local/bin/arabica --help
Set ARABICA_INSTALL_DIR for another directory or ARABICA_VERSION=v0.1.1 to pin a release. Windows archives are on GitHub Releases ↗
Rust 1.88+
Build arabica-cli from crates.io. Cargo puts the arabica binary in ~/.cargo/bin.
cargo install arabica-cli --locked
arabica --help
Prefer the repository source? Build the workspace with Cargo ↗
Arabica uses named provider and model entries, with API keys stored separately. Start from the guided configuration example and add Blend routing when you need multiple models.
Open the configuration guide[providers.openai]
api_type = "open_ai_responses"
base_url = "https://api.openai.com/v1"
api_key_env = "ARABICA_PROVIDER_OPENAI_API_KEY"
[models.default]
provider = "openai"
model_id = "your-model-id"
[blend]
default_policy = "coding"
[blend.policies.coding]
version = 1
default_model = "default"