Run a Rust-native coding agent from your terminal.

Install Roder, open it in a repo, and give it real work: inspect files, edit code, run commands, call providers, resume sessions, and expose the same runtime through a local app-server.

Install Roder Read install docs

Terminal UI and non-interactive `roder exec`
Provider-neutral model labels like `openai/gpt-5.5`
Local thread persistence, replay, and app-server control

Ghostty terminal showing Roder building a small Patchboard static site in a temporary repo — Real demo captured in Ghostty from `~/tmp/roder-cli-product-demo`.

Install Roder

The installer downloads the latest release archive, verifies its SHA-256 checksum, and writes `roder` to `~/.local/bin` unless you override the destination.

Quick install

Linux and Apple Silicon macOS

curl -fsSL https://github.com/RoderAI/roder/releases/download/latest/install.sh | bash

Install with Homebrew

Roder Homebrew tap

brew install RoderAI/tap/roder

Build from source

For local development

git clone https://github.com/RoderAI/roder.git
cd roder
make install

What the CLI gives you

Roder is the local product surface for the Rust harness: terminal UI, one-shot automation, provider setup, saved sessions, and the embedded control plane.

Work in any repo

Open Roder in a project and it attaches the run to that workspace, local config, tools, and saved thread state.

Use interactive or headless mode

Start the terminal UI with `roder`, or run a one-shot task with `roder exec` when scripts, evals, or CI need plain output.

Resume the exact run

Threads are saved locally, including transcript items, workspace metadata, provider/model selection, and usage.

Connect clients to one runtime

The embedded app-server gives TUI, desktop, IDE, ACP, and eval clients a shared JSON-RPC control plane.

Screenshots from a real Roder session

These captures use Ghostty. The build demo asked Roder to create a tiny static product card in `~/tmp`, then resumed the saved thread and checked provider/setup commands.

Real build demo screenshot — **Real build demo**Captured from Ghostty after a Roder run in `~/tmp/roder-cli-product-demo` created `index.html` and `styles.css`.

Resume saved work screenshot — **Resume saved work**The same thread reopens with the saved transcript, model, workspace, and input box ready for the next instruction.

Provider and exec surfaces screenshot — **Provider and exec surfaces**Auth status, media providers, and `roder exec --help` are plain CLI surfaces for scripting and setup checks.

Control plane screenshot — **Control plane**`roder app-server --listen stdio://` exposes the same runtime to trusted local clients and protocol adapters.

Use the model and provider you prefer.

Roder model labels use `provider/model`, and the default distribution includes first-party provider integrations plus custom OpenAI-compatible provider config.

OpenAICodexAnthropicGeminiVertex AIxAIOpenRouterFireworks AICursorClaude CodeKimi CodeRoder Cloud

Example config

provider = "synthetic"
model = "syn:large:text"

[providers.synthetic]
api_key_env = "SYNTHETIC_API_KEY"
base_url = "https://api.synthetic.new/openai/v1"

Interactive when you want it. Scriptable when you need it.

Use the full terminal UI for live development sessions. Use `roder exec` for repeatable tasks, eval harnesses, and automation. Start `roder app-server` when another local client should drive the same runtime.

roder
roder resume <thread-id>
roder exec --mode bypass "inspect this repo"
roder exec resume --last "continue"
roder app-server --listen stdio://
roder media providers

Install Roder and start in your repo.

One command gets the CLI onto your path. From there, run `roder` in a workspace or use `roder exec` for direct terminal automation.

Install Roder