OneManCompany (OMC) runs a simulated company of AI employees from a browser. You act as the sole human, and an org chart of agents - an executive assistant, HR, a chief operating officer, and a chief sales officer, plus hired engineers, designers, and QA - receives a task, breaks it down, executes it, and returns work for approval. It sits above coding agents rather than replacing them, orchestrating hierarchical teams built on those underlying tools.
The runtime is event-driven and separates a Vessel from a Talent: the Vessel is the execution container that defines retry logic, timeouts, tool access, and communication protocols, while the Talent is the capability package carrying skills, knowledge, personality, and specialised tools. Founding employees ship with the system; further employees are hired from a community-scored Talent Market and onboarded through an interview flow.
Work moves through company mechanics rather than a flat task queue: hierarchical task breakdown with quality gates at each level, multi-agent meetings with generated meeting reports, cross-functional delegation, structured handoffs, versioned project iterations, performance reviews with probation and promotion tracks, and one-on-one coaching sessions that persist as work experience for the employee. Per-project token usage and cost are tracked in real time.
Founding employees run in one of three execution modes, switchable in settings: a built-in hosted agent that calls models through OpenRouter, Claude Code with a Claude subscription, and an experimental OpenClaw mode with multiple model backends. The system needs Node.js 18 or later and Git, serves at http://localhost:8000, and sets up its Python toolchain on first launch. Roadmap items include more built-in tools, selectable frontend themes, additional model provider options such as local Ollama, Azure OpenAI, and AWS Bedrock, and more efficient multi-agent handoff.
Features
- Browser office: a pixel-art view of the company where employees, tasks, and meetings are visualised in real time
- Hierarchical delegation: the executive assistant routes work, the COO dispatches subtasks, and results pass through review gates
- Multi-agent meetings: employees call each other into synchronous discussions, with auto-generated meeting reports
- Talent Market: a community-verified marketplace of AI employees that HR searches, interviews, and onboards
- Vessel and Talent split: execution container and capability package are separate, with harness protocols between them
- Coaching and reviews: one-on-one sessions become lasting work experience; quarterly reviews drive probation, promotion, or termination
- Iteration management: structured version history per project, with rollback to an earlier version
- Execution modes: built-in hosted agent via OpenRouter, Claude Code, or experimental OpenClaw
- Cost tracking: per-project token usage and spend recorded per run
