Aegis Orchestrator
Zaru CLI

Installing the Zaru CLI

Build the zaru program from source, choose a model, and store a provider key.

There is no release to download. You build zaru from its repository with Cargo.

Requirements

  • Linux. The program is built and tested on Linux only. It uses Unix-only APIs, so it does not build on Windows. macOS is not tested.
  • Rust 1.98.1. The version is pinned in the repository's rust-toolchain.toml, and rustup installs it on the first build.
  • A C compiler (cc). Some dependencies compile C code.
  • Network access to crates.io for the first build.
  • A model to talk to: a Google Gemini API key, an Ollama server, or any server with an OpenAI-compatible chat completions API.

Build and install

git clone https://github.com/100monkeys-ai/zaru-cli.git
cd zaru-cli
cargo install --path crates/zaru-cli --locked

This builds the zaru binary and copies it to ~/.cargo/bin. Check that it runs:

zaru --version
zaru 0.0.0
  zaru-core 0.0.0
  zaru-tui 0.0.0
  zaru-notes 0.0.0
  zaru-seal 0.0.0
  zaru-aegis 0.0.0

The lines under the first name the crates the binary is built from. To build without installing, run cargo build --locked; the binary is then target/debug/zaru.

Choose a model

Zaru has no default model. You name one in ~/.zaru/config.toml, a file you create yourself. Each section below is a complete first configuration.

Gemini

Store your API key. It is read from standard input: paste it, press Enter, then press Ctrl-D.

zaru providers keys add gemini
stored a `gemini` key under the alias `provider.gemini`.
  the value is sealed and is not printed by any command.

Then name the model in ~/.zaru/config.toml:

[model]
default = "gemini-3.6-flash"

With a stored Gemini key and no provider.default.kind, Zaru uses Gemini for the default model.

Ollama

No key is needed. Start the server, then write ~/.zaru/config.toml:

[model]
default = "llama3.2"

[provider.default]
kind = "ollama"

Zaru connects to http://localhost:11434. Set provider.ollama.endpoint if your server is elsewhere.

An OpenAI-compatible server

Give the base URL; Zaru adds /chat/completions to it. If the server's /models list does not give the model's context window, give it in tokens too:

[model]
default = "your-model-name"

[provider.default]
kind = "openai-compatible"

[provider.openai_compatible]
endpoint = "http://localhost:8080/v1"
context_tokens = 32768

The kind is spelled openai-compatible and the configuration table openai_compatible. If the server needs a key, store it with zaru providers keys add openai-compatible; it is sent as a Bearer token.

Where stored keys live

Stored keys are kept in ~/.zaru/credentials.json, readable only by you, encrypted with AES-256-GCM. The encryption key is kept in your operating system's keyring.

If the machine has no keyring (a server, WSL, a container) and a task needs a stored key, zaru says so and stops. Set ZARU_CREDENTIAL_KEY to 64 lower-case hexadecimal characters in your shell profile, for example the output of openssl rand -hex 32, and keep that value: without it, the keys you stored cannot be read again. A task that needs no stored key, such as one for Ollama, goes on without it and says that the stored keys could not be read.

No command prints a stored key's value. zaru providers keys lists which providers have a key, and zaru providers keys rm <kind> deletes one.

Run your first task

From the directory you want it to work in:

zaru "read src/main.rs and tell me what it does"

Running a task explains what it prints, and Permissions what it asks.

On this page