Restless has introduced a new open-source platform designed to run entire companies using a structured AI harness. Unlike typical single-threaded chat interfaces, Restless assigns specific roles to agents, such as Production Lead, Build/QA, and Research, allowing them to coordinate complex workflows autonomously. The tool is currently in development preview and is self-hostable on macOS, Linux, and Windows with WSL2, requiring Docker for operation.

Structured Roles Replace Chat Chaos

The core innovation lies in treating AI agents as employees with defined authorities rather than just chatbots. The platform features roles like 'Exec' for company priorities, 'Marlow' for production leads, and 'Ines' for build and QA tasks. This structure ensures that agents handle routine choresβ€”such as sprite exports, device checks, and bug fixesβ€”while human users retain final judgment on creative direction and external publishing. The system enforces authority limits via code, ensuring agents cannot expand their own mandates or access raw secrets without permission.

Real-World Application with FrankenSurf

To demonstrate viability, the team showcased 'FrankenSurf,' an open-source project with 100 GitHub stars that is actively maintained by a Restless company. The AI agents write code, ship releases, and handle marketing for the project. The interface displays a persistent Linux machine where agents execute tasks like packaging demo builds and running overnight soak tests. Users can track events in real-time, seeing exactly which agent handled which task, such as 'Theo AI' comparing launch pricing or 'Camille AI' drafting press kits.

Pricing and Model Flexibility

Restless supports multiple model providers, including OpenAI, Anthropic, and OpenRouter, allowing each agent to use a different model based on cost and capability needs. The platform offers a free Core tier for self-hosting, while hosted plans are in waitlist stages. The Founder tier costs $49/month for two companies with $20 of AI use included, and the Scale tier is $149/month for five companies with $75 included. This pricing model shifts the burden of AI usage costs to a predictable monthly fee, with options to top up at cost.

Key Takeaways

  • Role-Based Autonomy: Agents are assigned specific company roles (Lead, Worker, QA) with strict authority boundaries to prevent scope creep.
  • Persistent Infrastructure: The system runs on a persistent Linux environment with Git and browser access, allowing agents to maintain state across sessions.
  • Human-in-the-Loop: Critical decisions, such as publishing to outsiders or approving final demos, are flagged for human review, while routine tasks are automated.
  • Model Agnostic: Users can mix and match AI providers (Claude, Codex, Gemini) per agent, optimizing for specific tasks like research vs. coding.

The Bottom Line

This is a serious attempt to move beyond 'prompt engineering' into 'agent orchestration.' While the 2-point HN score suggests early-stage traction, the architectural focus on persistent state and enforced authority limits addresses the biggest pain points of current agentic workflows. If the hosted beta delivers on the promise of unattended overnight builds, this could become the standard operating system for solo developers running AI-native companies.