OverlordiusWindows and Linux app
From planning to delivery, AI agents working on your project.
Describe what you want to create or improve. Overlordius organizes the work with agents that take on roles such as planning, development and review, following your project's instructions. Tasks can run in a queue or in parallel. Multiple agents can also work at the same time within a single task. You approve plans and follow progress in a visual interface. The hub coordinates execution, checks changes and integrates the results into your project, bringing you in when a decision needs your attention.
Overlordius is completely free, with all current features included.
BYOK (Bring Your Own Key): connect your own API keys. Providers charge for model usage according to your account.
Product stage
Closed alpha
Overlordius is not available to the public yet. Before opening access, we are checking with OpenAI and Anthropic whether sign-in can be used with ChatGPT and Claude subscriptions.
All current features will remain free. If integration with ChatGPT and Claude subscriptions is approved, it will also be free in Overlordius. Register your interest to follow access opening.
See what's next on the roadmap01
The problem
Running several coding agents at the same time is manual work today:
- one terminal per agent
- branches and worktrees created by hand
- assigning tasks and tracking dependencies
- approving permissions all the time
- checking what each one did
- merging the work and resolving conflicts
You end up babysitting the agents, and a project without good instructions makes them make more mistakes.
02
The proposal
Overlordius takes you out of the middle: the agents verify, escalate and recover on their own. You organize plans in queues to run them in sequence, without starting each stage manually. Depending on the context of each slice, the hub continues in the same chat to reuse what has already been discussed or automatically opens a new chat for the next stage.
You're only called for what is your decision:
- product
- taste
- cost
03
What it does today
Click any feature to open or close all descriptions.
An agent reads the project, writes the plan and proposes the tasks. A plan can split into multiple worktrees and slices (stages of execution). Slices can continue in the same chat, reusing its context, or run in a new chat. Each worktree has its own branch, and the hub commits each step.
The hub selects models by kind of work and difficulty, among those connected, active and allowed in the project. You can adjust the ladders and spending limits per AI.
When a task finishes, it goes back to the target branch only after passing the project's verification. A conflict or a failed check opens an agent that resolves it.
Organize plans in queues to run them in sequence. Plans and tasks wait their turn without you having to watch each one start.
An agent creates or adapts project instructions (
AGENTS.mdand related files), organized by area. Harness verification proposes updates with evidence for your review.An agent talks through the idea, reads the specifications and proposes the project brief, the harness and the base plan.
Check project and plan progress, understand blockers and ask the assistant to act. It resolves what it can and asks for your decisions when needed.
Track usage and spending for each AI connected through an API key.
Connect AIs with your own API keys: Anthropic, OpenAI, DeepSeek and OpenRouter. Keys and projects stay on your computer.
Track issues by project and plan, with severity, evidence and resolution. Verification or integration failures can trigger repair agents.
Each delivery records changes, criteria, checks and concerns tied to commits. See what was verified and what was only reported by the agent.
Compare candidates with a baseline and see the hub's recommendation, blockers and unassessed changes before choosing what to ship.
Prepare a package from a selected commit in an isolated worktree, tracking progress, artifacts and checks. Preparation preserves your work and requires a separate action to publish or install.
Extend agents with browser and language tools, with requirements and permissions managed by the hub.
Connect tools to the hub through local APIs in Messages, Responses and Chat Completions formats, using the selected AI.
04
Who it's for
For people using AI agents to build software who want to coordinate work with less manual supervision, from an initial idea to improvements on an existing project.
Companies and teams
Organize agent work by project, with defined instructions, plans and verification criteria. Follow parallel tasks and consult delivery history to understand what changed and what was verified.
Solo developers
Bring an idea to life, improve a product or handle fixes without coordinating each agent in a separate terminal. Use agents to plan, implement and review, choosing models to suit the work and your budget.
People who want more freedom in their day
Queue plans and follow progress in the interface without starting each step by hand. Reduce interruptions to handle other tasks, then catch up through findings and delivery reports.
You still define the goals, approve plans and evaluate results. The hub coordinates the agents and brings you in when it needs a decision.
Follow the opening
Want to use Overlordius?
We are in closed alpha. Leave your email to express interest and receive news about access opening. Downloads and account creation remain invitation-only.