OpenEng AI quickstart

Sign in once, learn the Console in five minutes, publish your first config, and start the engine with your OpenEng key — then run it from the Web Console at app.openeng.app.

The steps, in order:

1. Sign in to the Console.
2. Find your way around the Console.
3. Publish your first config: the engine serves nothing until you do.
4. Create your OpenEng key.
5. Install and start the engine with that key.
6. Open the Web Console, pick your engine and config, and start working.

1. Sign in

You have one OpenEng account, and every surface signs in through auth.openeng.app with Google — the Console, the Web Console, the website, the billing site, and the five engineering apps all share it.

Open console.openeng.app. If you're not signed in you land on the login page; Continue with Google completes the sign-in and returns you to the page you were heading for.

2. The Console layout

The Console is the authoring side of OpenEng AI: you define things here, publish a version, and the engine on your machine runs it. The shell has three parts:

The left nav puts the config first and groups everything else by intent:

The sidebar collapses to icons with the Collapse button at its bottom (remembered per browser). On phones it becomes a slide-in drawer behind the menu button. The Console tour explains every page.

The command palette — press ⌘+K (Ctrl+K on Linux and Windows) anywhere, or click the Search or jump to… box in the header. It jumps to any page and carries creation actions (New model, New agent, Register cell deployment, New config, …) plus Publish a config and Import a library. Esc closes it.

The Overview page is a working dashboard: a first-run checklist you can dismiss (add a model, create an agent, create and publish a config), your configs and whether they're published, counts of your building blocks, your plan with its meters, recent activity, and the Deploy strip. There are no usage/telemetry charts — the Console manages configuration and your fleet, not analytics.

Light & dark theme

Account → Preferences has a two-state Light/Dark switch (light is the default). The choice is stored to your account — one switch renders everywhere you sign in, on every device. Locally it also applies before first paint, so there's no flash; after sign-in the server-stored value wins over the local cache.

3. Publish your first config

The engine runs only what you publish, and until your account has a published config it refuses every request. A new account starts with no config.

1. Build something — e.g. an agent: Agents → New agent, give it a name and a system prompt, and press Save in the editor header (Agents). Add a model on the Models page too, or rely on the embedded defaults.
2. Open Configs (guide). On a new account the page shows Create your first configuration: type a name, such as workstation, and press Create. The guided config builder opens.
3. Walk the steps in the rail. Most are optional: the functionalities in Settings start switched on, and model groups you leave unset in Models use the engine's embedded defaults. The component steps list only your own items, so the agent you just saved is there to tick.
4. On Review & publish, keep the suggested version (1.0.0 for a first publish), add a note, check the changes and the validation list, and press Publish v1.0.0.

The version is published immediately. Everything you publish later works the same way: you bump the version, and old versions stay available.

4. Create your OpenEng key

The engine authenticates with an OpenEng key, which is an API key you issue in the Console:

1. Open Deploy → API keys and press Issue key.
2. Give it a label (for example, the machine it will run on). Scopes are optional, and they're recorded but not enforced yet: every key has full account authority.
3. Copy the key. It's shown exactly once; the Console stores only a hash of it.

API keys need Pro: on the Free plan the API keys page explains what Pro unlocks instead of listing keys, and on Pro Issue key is disabled, with the reason, once you reach your plan's key limit. Each key runs one engine at a time (API keys & sandboxes).

5. Install and start the engine

If you haven't installed the engine yet:

curl -fsSL https://openeng.app/engine/install.sh | bash

The installer puts the openeng binary in ~/.local/bin, checks it against its published SHA-256 checksum when it can (it prints checksum verified; see Check a download yourself), and downloads the default chat model. It's published for Linux x86_64; on Windows, run it inside WSL (Install covers platforms, options and uninstalling). The same command is on openeng.app/download.

Then start the engine with your key:

openeng serve --key=<your OpenEng key>

# or keep the key in your environment
export OPENENG_KEY=<your OpenEng key>
openeng serve

The engine registers the key with the OpenEng API, which checks it and your Pro plan. It then applies your default config, downloads the models that config needs, and dials the OpenEng Broker. It opens no port on your machine; the Web Console reaches it through the Broker. There is no --oauth flag for this engine and no browser sign-in step. The engine reference lists every command, flag and environment variable, and what each startup message means.

6. Open the Web Console

Open app.openeng.app and sign in with the same Google account as the Console. The Web Console then walks you through three screens:

1. Select an engine — your account's engines, each marked online or with when it was last seen; pick one and press Continue. A newly started engine appears on its own. If none is connected, the screen shows the start command, openeng serve, to copy (it expects OPENENG_KEY in your environment).
2. Pick a configuration + version — only published configs are listed. With none published, you see the same "No configuration published" message the engine gives; publish one in the Console and the list updates on its own.
3. Warming up your configuration — the engine prepares everything the config needs, such as model downloads, with progress for each. Chat opens automatically when it's ready; Skip to chat lets you start while it finishes. If a required resource fails, the screen shows the failure with a Retry button.

If the config uses ${ENV} placeholders, give them values on the Web Console's Placeholder values screen for that config, or export them as environment variables where the engine runs; the environment wins.

How the engine picks up what you publish

The engine never authors configuration — it resolves what the Console published, from api.openeng.app.

Three things worth knowing from day one:

• You choose the version per session. The Web Console's picker lists every published config and version, and refreshes while you pick, so a version you just published shows up within seconds. The engine fetches a version the first time a session uses it.
• Versions are frozen — a published version never changes, so a session keeps running the version it picked. To use a newer publish, pick the newer version. Each version also has a copyable pinned id, <id>@<version>, on its config's version timeline.
• Requests that name no config use the engine's default: the config given with --default or OPENENG_DEFAULT_CONFIG, otherwise the config you own named default, otherwise the first config you own (The engine).

Plan notes

The Free plan includes signing in, the Console, authoring models, agents, templates, MCP servers and cells, and building and publishing configs. Pro adds schedulers, API keys, sandboxes, storage points, the Research and Device Use functionalities (Device Use also needs an engine build with the X11 device backend, which the published engine doesn't include yet), Roles, Human-in-the-Loop and an included planning credit — and running the engine requires Pro. Pages gated to Pro explain what it unlocks, and the nav wears a lock badge on Schedulers, Storage points, API keys and Sandboxes while the gate is closed. Pro launches in Q4 2026; see Plans & billing for the current state of AI pricing. As of September 25, 2026 the OpenEng AI services aren't live on any plan (see the status note at the top of this page).

Troubleshooting

The Console says the platform API isn't reachable. The banner reads "The platform API isn't reachable right now — your session is fine; data will load once the API is back." The Console retries automatically, and Retry now tries at once. Editors mirror your draft locally, so nothing is lost while it's down.

"Plan unverified" on a locked page or the Account page. The plan couldn't be confirmed (a transient network or sign-in hiccup), so the Console shows the Free view defensively. Press Refresh next to the notice — a paying user is never silently stranded.

The engine won't start. Read the line it printed: a missing key, a rejected key, a missing Pro plan and a key that already runs an engine each have their own message (engine troubleshooting).

The Web Console says "No configuration published". Publish a config in the Console (step 3); the Web Console picks it up on its own.

The engine doesn't run what I just published. Check the three usual suspects: the same account in the Console and the Web Console, the version picked for your session (pick the new one), and whether the item is disabled (a "Not publishable" chip in its editor means the engine deliberately skips it).

Sign-in loops back to the login page. Third-party-cookie or storage blockers can break the redirect round-trip. Allow storage for console.openeng.app and auth.openeng.app, then retry.