Prime Intellect has released Prime Agent, an open-source coding and research agent framework designed for long-running, autonomous workflows. The architecture is built upon two primary technical abstractions: the Recursive Language Model (RLM) and the Continual Harness. The RLM treats context as variables and utilizes a persistent Python REPL as a built-in tool, allowing for programmatic subagent invocation via the rlm.spawn() method. This enables parallel or background task execution where child agents can communicate directly without user mediation. The Continual Harness serves as a durable state repository for supplemental prompts, memories, and reusable subagent specifications. Users can employ the /refine command to apply evidence-backed updates to this harness, which maintains an immutable base system prompt while allowing for recorded refinement history and rollbacks. For extended operations, the system utilizes a daemon-backed architecture that preserves Python REPL state, schedules, and goals across terminal disconnections. Advanced autonomy is managed through bounded modes, allowing users to configure specific turn, token, and time budgets alongside custom quality gates. While the system provides lifecycle isolation through worker and kernel processes, it does not function as a security sandbox, executing model-generated Python and shell commands with the user's local permissions.