Prime Agent
An AI agent system implemented in the Rust programming language, designed with an emphasis on performance and memory safety. It is intended for production environments that require high concurrency and long-running operation.
Timeline (last 90 days)
A discussion about rewriting Prime Agent in Rust appeared on Hacker News, receiving 15 points and 3 comments.
The Prime Agent team ran a Factorio-style experiment that streamed continuously for seven days, mobilized 633 agents, and consumed 23 million output tokens, finding that the model was able to continuously advance the tech tree without getting stuck.
Prime Agent achieved approximately 78% on ARC-AGI using the GPT series and 95.5% using Opus.
With an appropriate Harness, Prime Agent achieved 95% on ARC-AGI.
pstack-claude supports multiple mainstream agents including Claude Code, Codex, Pi, OpenCode, Gemini, and Prime Agent.
Prime Agent achieves a positive feedback loop — the more it is used, the richer the skill library becomes and the higher the execution efficiency — by reusing a validated skill library.
Prime Agent supports spawning sub-agents to work in parallel, and when faced with complex tasks, it can decompose them and assign the parts to multiple sub-agents for simultaneous processing.
Prime Agent gained 2,655 stars on GitHub in a single day and topped the trending list.
Prime Agent represents the path of pursuing the upper bound of AI capability, while Semantica represents the path of upholding the lower bound of AI trustworthiness.
Prime Agent采用递归语言模型（RLM），不提供固定工具集，而是提供持久化的iPython内核，让模型通过编写Python代码创造工具
7 more timeline events
All Facts (16)
With an appropriate Harness, Prime Agent achieved 95% on ARC-AGI.
50%UnverifiedPrime Agent achieved approximately 78% on ARC-AGI using the GPT series and 95.5% using Opus.
50%Unverifiedpstack-claude supports multiple mainstream agents including Claude Code, Codex, Pi, OpenCode, Gemini, and Prime Agent.
50%UnverifiedPrime Agent supports spawning sub-agents to work in parallel, and when faced with complex tasks, it can decompose them and assign the parts to multiple sub-agents for simultaneous processing.
50%UnverifiedPrime Agent achieves a positive feedback loop — the more it is used, the richer the skill library becomes and the higher the execution efficiency — by reusing a validated skill library.
50%UnverifiedPrime Agent represents the path of pursuing the upper bound of AI capability, while Semantica represents the path of upholding the lower bound of AI trustworthiness.
50%UnverifiedPrime Agent采用递归语言模型（RLM），不提供固定工具集，而是提供持久化的iPython内核，让模型通过编写Python代码创造工具
50%UnverifiedPrime Agent不提供固定工具集，而是提供一个持久化的iPython内核，工具、技能和子智能体都是模型可直接编写运行的Python代码，这一设计称为递归语言模型（RLM）
50%UnverifiedPrime Agent的持续协作系统（Continual Harness）将提示词、习得技能和子智能体定义存储为可编辑状态，智能体能自行修改，通过Refine命令永久更新协作系统本身
50%UnverifiedDeepSeek Harness、Prime Agent、Pi和OpenCode四款开源工具都主张套壳与模型应该解耦
50%UnverifiedPrime Agent的持续协作系统允许智能体自行修改其提示词、习得技能和子智能体定义，实现自我优化
50%UnverifiedPrime Agent的Worker和内核进程使用本地用户全权限运行，而非沙盒环境，官方建议在隔离环境中使用
50%UnverifiedA discussion about rewriting Prime Agent in Rust appeared on Hacker News, receiving 15 points and 3 comments.
50%UnverifiedThe Prime Agent team ran a Factorio-style experiment that streamed continuously for seven days, mobilized 633 agents, and consumed 23 million output tokens, finding that the model was able to continuously advance the tech tree without getting stuck.
50%UnverifiedPrime Intellect开源了一款能在运行中重写自身指令的智能体Prime Agent
50%UnverifiedPrime Intellect于2026年8月初以MIT许可开源了Prime Agent智能体
50%