OverlordiusHow to use
Go from signing in to your first delivered plan in six steps.
Create your account with an invitation code in the app or on the website. Sign in on the website to download, then use the same account in the app.
Your first plan
Sign in or create an account
Open Overlordius and sign in. To create an account, enter your invitation code, email and password, then confirm your email. GitHub and Google sign in to accounts that already exist.
Check your AI connections
In “Connections”, save your Anthropic, OpenAI, DeepSeek or OpenRouter API key and click “Test”.
Create a project
Choose “New project”. Open an existing folder or start from scratch with an agent.
Set up the project harness
Choose “Configure with an agent” (recommended) to set up the project's harness: the instructions and tools that guide the agents. Review the agent's proposal before clicking “Finish”.
Describe and approve a plan
Choose “New plan” and describe what you want. The orchestrator proposes tasks organized into worktrees and slices (stages of execution); review and approve the plan.
Follow the work through integration
Follow progress in “Overview”. Use queues to run plans in sequence. The hub only calls on you for decisions. When the work is complete, each task integrates into the branch you chose.
Models by kind of work
Each plan step names its kind of work and difficulty. The hub picks the cheapest model good enough for it from those connected, active and allowed in the project. The recommended ladders work without setup; adjusting them is optional.
Four kinds of work
- Coding — implement features, fix bugs and refactor code.
- Design — work on screens, visual design and user experience.
- Planning — define scope, architecture and steps of execution.
- Review — check changes and look for problems before integration.
A ladder for each kind
Each ladder has four difficulty levels: Simple for a fully specified edit, Standard for a clear feature or fix, Hard for work across modules or a bug that needs investigation, and Hardest for open architecture decisions or especially sensitive work. Each level lists model and effort pairs.
In Settings › Hub › Models by kind of work, use “Edit ladder” to adjust the pairs or restore the recommendations. The proposal shows the chosen pairs; approval saves them, so later ladder changes do not move an approved plan to other models. A failed step retries one level higher with another pair; a step pinned to a specific pair pauses instead.
Connected and active models
Connecting an AI makes its models with a model card active by default. A model without a card stays off until you choose “Activate model”; it also needs a card before the hub can select it automatically. Manage these choices under “Active models” in Settings › Hub.
“See recommended” explains whether a recommended model is missing because its AI is not connected, the model is off or it is outside your ladder. If no suitable model is available, the step waits and the hub says what needs fixing. Turning a model off blocks future project turns, including manual choices; a running turn finishes.
Restrictions for a project
A project's Configuration can narrow the allowed models and set routing rules in its harness. It cannot add a model that is disconnected or inactive in the hub.
If a project cannot use a model, check its restrictions as well as the AI connection, Active models and the relevant ladder.
Orchestrator and task models
The orchestrator uses the AI and model you choose in the New plan form. It proposes and coordinates the work; the ladders choose the models that execute the task steps. Choosing an orchestrator does not assign that model to every task.
The conversation that starts a project uses the AI and model chosen when creating that project. You can therefore use one model to organize the work and different models to carry out its steps.