outrig runs an LLM agent on the host and connects it to MCP servers inside a podman-managed
container, so tools like filesystem and shell access stay inside the sandbox you've set up.
Crate names matter: depend on outrig when embedding the library, and install
outrig-cli when you want the command-line tool that provides the outrig binary. Both
crates publish from this repository.
You define the sandbox in your repository: a Dockerfile for the container, an
.agents/outrig/config.toml describing which MCP servers run inside it and which LLM the agent
talks to. Then you run outrig run from your shell, and outrig drops you into a stdin/stdout REPL
where you talk to the agent.
The agent layer is the Rig crate -- outrig's job is to run Rig in the host process and give it a container-isolated MCP tool set.
Release action: reserve or publish both
outrigandoutrig-clion crates.io before removing this note from the first public install instructions.
Install the CLI package to get the outrig command:
$ cargo install outrig-cliFor library use, depend on the outrig crate from your Rust package.
$ outrig run
[outrig] agent: coding (model: fast / provider: openai / gpt-4o-mini)
[outrig] container outrig-20260502T103412-3f2a started
[outrig] mcp servers: fs, shell
> what's in this repo?
This repo is a Rust project named "outrig". The top level contains Cargo.toml,
src/, and doc/. src/ has lib.rs and main.rs...
> ^D
[outrig] session 20260502T103412-3f2a endedThe full docs live in doc/ and render as an mdbook. The latest trunk
build is published at https://tgockel.github.io/outrig/. To preview locally from the
project root:
mdbook serve # live preview at http://localhost:3000
mdbook build # static HTML in target/book/Start at the Introduction or jump to the Quickstart.
See CONTRIBUTING.md for local checks, the e2e test invocation, and docs build instructions.
Licensed under the Apache License, Version 2.0. See LICENSE.