KandaQuantum, a Tokyo-based company, has formally released Fuga 5000, an orchestration platform that enables developers to direct large scale AI agent swarms from a single Mac. The system coordinates up to 5,000 agents per day by connecting with existing subscriptions for Claude Code, Codex, Gemini, Grok, and Kimi. This launch marks an expansion from earlier closed releases and positions Fuga 5000 as one of the largest commercial agent orchestration tools available.
Scaling Agent Swarms from One Machine
Fuga 5000 supports five hierarchical layers, ranging from request handling and goal management to task-level coordination, main agents, and subagents. In a recent measurement on October 16, 2026, a MacBook Pro with M4 Max and 128GB of memory directed 5,441 agents across 349 tasks, an average of roughly 15.6 agents per task. Because inference runs on vendor clouds, the local Mac mainly handles agent startup, management, and command execution while code remains on the local machine.
Automatic Assignment and Real-Time Scheduling
The platform draws on a benchmark ledger that ingests public results from Terminal-Bench, SWE-bench, and Artificial Analysis in real time, matching each task to the most suitable model and vendor mix. Users can set vendor allocation preferences through a pie chart, and the system displays assignments on a Gantt chart covering planning, parallel implementation, integration, reporting, and independent inspection. If one subscription reaches its usage limit, work is automatically shifted to another vendor account.
Subscription-Based Cost Efficiency
Fuga 5000 operates through official agent command line interfaces using each developer's flat-rate subscriptions rather than API usage-based billing. KandaQuantum reports that in September 2026, running all 89 Terminal-Bench 2.1 tasks with around 800 agents would have cost about US$475 under API billing, but the same work fit within one day of the flat plan at roughly US$47. The service itself costs US$200 per month, with an introductory first month price of US$150, excluding separate vendor subscriptions.
Safety and Oversight Features
To manage risks across thousands of parallel agents, Fuga 5000 monitors operations and pauses agents when dangerous actions are suspected, such as modifying files outside a task area or deleting protected files. Users can review the stopped action, the preceding tool calls, and the agent's recent statements before choosing to continue or stop. The system also generates requirement definition documents and final reports automatically, reducing manual documentation effort.
Performance Metrics and Agent Activity
The Agent activity dashboard tracks actual working hours, allocated agent counts, and token consumption by day and month across vendors. It converts agent labor into an estimated human cost figure using a provisional rate of 250,000 yen per hour and visualizes cost performance in scatter plots. The company also reports that agent orchestration scale has roughly doubled each month, from about 1,000 agents in August to over 2,000 in September and around 5,000 in October.
Fuga 5000 represents a notable step toward practical, high-volume AI agent orchestration for individual developers and small teams. It is offered through the KAMUI and KAMUI OS product line, which reached 100 million yen in annual recurring revenue within four months of its March 2025 release. The platform will be available first on macOS, with Windows and Linux support planned for the future.