Oriel

Try it live ↗ · the full UI in your browser, mock data, no install.

[Oriel Studio (light)] [Oriel Studio (dark)]

Drive it with AI

Point any MCP client at Oriel and ask in plain English. It calls the same validated, secret-masked tools the UI does, and it can chain several to actually answer a question instead of firing just one:

you: why does api keep restarting?

assistant: reads container.list, container.inspect, then the logs. It's exiting 137 (OOM): the container is capped at 256 MB and the Node heap runs past it. Raise the memory limit or fix the leak in worker.js.

oriel mcp        # stdio MCP server for Claude, Cursor, or a local LLM

No model ships in the binary; your client brings its own. Reads run anytime, and destructive actions stay locked behind a grant. Setup & tool list ↓

Install

Homebrew (macOS & Linux):

brew install ParadoxInfinite/oriel/oriel

Script (detects your platform and verifies the checksum):

curl -fsSL https://raw.githubusercontent.com/ParadoxInfinite/oriel/main/install.sh | sh

Add -s -- --edge to track pre-releases (newest builds first), or -s -- --uninstall to remove it:

curl -fsSL https://raw.githubusercontent.com/ParadoxInfinite/oriel/main/install.sh | sh -s -- --edge

⚠️ Piping a script to sh runs code from the internet on your machine. Read install.sh before you run it, or download a binary from releases instead.

go install github.com/ParadoxInfinite/oriel@latest   # with Gomake build                                           # from source (builds + embeds the UI)

Run

./oriel            # opens http://127.0.0.1:4321

Flags: --port <n> (default 4321), --no-open. Run on login: ./oriel service install (launchd / systemd; also status, uninstall).

Needs any Docker Engine–compatible runtime + the docker CLI. Colima is first-class (adds VM start/stop); Docker Engine, OrbStack, Rancher/Docker Desktop, Podman, and remote daemons also work (docs/DAEMONS.md). For remote access over a private network, see docs/REVERSE-PROXY.md; read the security note first.

Features

• Containers: live CPU/mem, exit codes, bulk actions, streaming logs, full inspect.
• In-browser shell: open an interactive terminal into any running container, straight from its details drawer.
• AI control (MCP): any MCP client drives Docker/Colima through the same validated, secret-masked tools, with destructive actions behind a grant. More ↓
• Images: pull with registry search, prune, one-click tag.
• Compose: manage stacks, plus discover & deploy projects from disk.
• Dashboard: CPU history, memory, disk, uptime/outage tracking.
• Command palette (⌘K): fuzzy-run any action or jump to any view.
• Editions, themes & languages: swap the whole UI (Studio, or drop in your own), light/dark/system, custom accents, and a translatable interface (English ships; other languages load on demand).
• Light & live: ~15–30 MB RAM, one SSE stream (no polling), checksum-verified self-update.

How it compares

The usual ways to run containers on a Mac or Linux box, and where Oriel fits. (Figures drift; treat as a snapshot.)

It's the only one here an AI can drive directly, through the same checks the UI gives you. Reach for it over Docker Desktop / OrbStack (no paid license, bundled VM, or menu-bar app), over lazydocker (a real graphical UI, not a terminal one), or over Portainer (a binary you run for yourself, not a server to deploy and lock down).

For the full breakdown, including where Oriel loses (Windows, Kubernetes, multi-host), see the exhaustive comparison in the live demo.

AI control (MCP)

oriel mcp runs Oriel as a Model Context Protocol server, so an MCP client (Claude Desktop, Claude Code, Cursor, a local LLM) manages your Docker/Colima in plain English, headless, with no GUI needed. Same tools, same guardrails:

• Secrets stay masked. container.inspect and container.logs redact secret-shaped values before they reach a model; an MCP client never gets fully-raw env or logs (the "off" setting applies only to the local UI). Log redaction is best-effort over free-form text.
• Destructive actions are locked until you open a short, time-boxed window (oriel ai allow-destructive --for 15m). Reads always work; remove/prune don't, until you say so.
• Every action is logged. An audit log records each tool call an assistant makes: what ran, with which arguments, and when (Settings → AI activity). Your own clicks aren't.
• No model in the binary. Your client brings the model, so Oriel stays vendor-neutral.

Point any MCP client at it. With Oriel installed:

{ "mcpServers": { "oriel": { "command": "oriel", "args": ["mcp"] } } }

Or with no install, via the published image (Linux hosts; mounts the Docker socket):

{ "mcpServers": { "oriel": { "command": "docker", "args": ["run", "-i", "--rm", "-v", "/var/run/docker.sock:/var/run/docker.sock", "ghcr.io/paradoxinfinite/oriel"] } } }

Claude Code: claude mcp add oriel -- oriel mcp. Setup, HTTP, scoping, and the full tool list: docs/MCP.md.

Editions & themes

The UI is a swappable plugin on a stable platform SDK: Studio (light/dark/system, custom accents). Recolor it, or drop in your own edition (see docs/THEMES.md).

Security

Out of the box Oriel has no login, and driving Docker is effectively root on the host. An optional bearer token gates remote and MCP-over-HTTP access (off by default), but the safe default is local use, or a private network only (Tailscale, ZeroTier, WireGuard, and the like). Never the public internet. Full trust model: SECURITY.md.

FAQ

Is there a GUI for Colima? Yes. Colima ships CLI-only by design, and Oriel is the browser UI it never shipped. It also drives Docker Engine, OrbStack, Podman, and remote daemons.

A free Docker Desktop alternative? Yes. Apache-2.0 licensed, no license fees, no account. Point it at any Docker-compatible engine.

Can an AI manage my containers? Yes. Run oriel mcp and point any MCP client (Claude, Cursor, a local LLM) at it. It gets the same validated, secret-masked tools as the UI, and destructive actions stay locked until you grant them.

How is it different from lazydocker? lazydocker is a terminal UI; Oriel is a graphical browser UI with dashboards, streaming logs, registry search, Compose discovery, and themeable editions.

Footprint and platforms? ~15–30 MB RAM, ~13 MB binary, no Electron. macOS (Apple Silicon + Intel) and Linux (amd64 + arm64).

Develop

make dev + make dev-web (Vite hot reload), make test. See CONTRIBUTING.md.

License

Apache-2.0 © The Oriel contributors. See NOTICE.