Most AI agents are optimized to complete a task. openJiuwen's WorkSwarm is pushing toward something more ambitious: agents that improve themselves through real work. Three things that make it different 👇 1.⁠ ⁠Dual-Dimension RSI A continuous self-improvement loop — Evaluate → Analyze → Propose → Execute → Adopt — that improves both how the agent works (prompts, skills, tools) and what it delivers (papers, code, artifacts). 2.⁠ ⁠Persistent Sessions Keeps roles, decisions, and boundaries intact across hundreds of turns. In one test: 5 people, 189 turns, 8/8 cross-role conflicts caught. 3.⁠ ⁠Compute Affinity Optimizes resource usage — 22.8% lower HBM peak, 14.36% higher KV cache hit rate — making repeated optimization more efficient. Open source under Apache 2.0. GitHub 👉 #AI# #Agent# #openJiuwen# #WorkSwarm#