orka
Run LLM coding agents in isolated containers.
Mount what they need. Nothing else is reachable.
Filesystem isolation
Agents run inside Docker containers. They can only see the directories you explicitly mount in — not your home directory, credentials, or shell history.
Reproducible environments
The agent environment is defined by the image and your mounts, not by whatever happens to be in your PATH at the time. Consistent across machines.
Three runtimes
Supports pi, claude-code, and codex. Switch runtimes per invocation. Images are built and cached independently. Works with Docker, Podman, and nerdctl.
Presets
Named configurations in environments.yaml that mount volumes
and inject env vars. Stack multiple presets in one invocation.
File shadowing
Keep files and/or directories out of the agent’s context.
Files matched by an orkashadow file are replaced with empty
read-only stubs inside the container — visible but unreadable.
Common invocations
| Command line | ~/.config/orka/config.yaml |
|---|---|
orka | |
orka -f src/main.rs -f Cargo.toml | |
orka --preset rust | |
orka --scratchpad research | |
orka --tmp | |
orka --engine podman | engine: podman |
orka --runtime claude | runtime: claude |
orka --harness-version 1.2.3 | harness: "1.2.3" |
orka --no-browser | no_browser: true |