Plasma AI builds infrastructure for intelligence at scale, providing the foundations necessary for autonomous AI agents to perform complex, long-running tasks. As AI transitions from a single tool to an organized, coordinated workforce, Plasma AI creates the systems needed to manage these agents, ensuring they operate within defined boundaries with clear ownership, auditability, and shared knowledge. By moving beyond isolated agent sessions, Plasma AI enables organizations to treat fleets of agents as a reliable, scalable workforce. The company focuses on the practical needs of autonomous operations, such as budget management, task approval, and durable record-keeping, allowing for human intervention and control throughout the execution process.

Some of the key features are:

• Fractal: A recursive agent loop system that organizes agents into a tree hierarchy where each node operates within its own git worktree, memory, and budget to handle complex, multi-step tasks.
• Plasma Wiki: An indexed Markdown knowledge base that provides a structured, searchable context for agents, preventing redundant data loading and ensuring shared project context.
• Parallel Coordination: Enables the orchestration of agent fleets to execute work in parallel, with clear handoffs and reconcilable results.
• Durable Traceability: Maintains a permanent record that connects every result to the specific work performed and the decision-making process behind it.
• Operational Controls: Provides mechanisms for setting boundaries before autonomous work begins and enabling human oversight while processes are running.

Plasma AI operates by providing specialized infrastructure tools that integrate with existing agentic workflows. Their tools are designed to work with common coding agents like Claude Code or Codex, extending their capabilities through recursive loops and structured data storage. Developers install the Fractal or Wiki command-line interfaces to initialize environments within their existing repositories. Fractal handles the execution of multi-step work, while Wiki serves as the persistent memory layer for these agents. This architecture allows for complex tasks to be broken down into smaller, manageable sub-tasks that can be independently assigned and tracked within the system.

Some common use cases include:

• Large-scale refactoring: Using Fractal to automate complex code migrations that span multiple services and require iterative testing and verification.
• Automated knowledge management: Maintaining a project-wide wiki that agents can query to stay updated on architecture, conventions, and workflows without requiring manual context refreshing.
• System stress testing: Running recursive agent loops on a codebase to automatically identify, catalog, and fix bugs in a systematic, multi-phase process.
• Research campaigns: Executing multi-step experimental campaigns where each subtask is assigned to an agent node with a specific budget and goal.
• Autonomous service maintenance: Deploying agents to monitor, debug, and patch production services, with human-in-the-loop controls for safety and verification.