Prime Agent is an open-source tool that applies Recursive Language Models to coding and research tasks. It treats context as variables and uses recursive subagents as function calls within a persistent REPL environment. The system includes a Continual Harness to store durable state, memories, and skill descriptions for evidence-backed refinement.
- Uses RLM abstraction to treat prompts as variables for dynamic context management.
- Supports programmatic tool calling via recursive subagents in a persistent REPL.
- Continual Harness stores durable state including memories and skill descriptions.
- Enables small, evidence-backed updates to agent capabilities during sessions.
