Hello HN! I’m excited to introduce AI-gent Workflows, a new AI Agents platform featuring a local reasoning layer and a unique state machine design. This platform, which operates smoothly on mobile devices, allows for easy UI sessions akin to remote desktop connections. It addresses some noted design flaws in other frameworks by implementing “organic workflows” through a stateful flow graph, enabling detailed debugging of decision-making processes.
Large Language Models (LLMs) effectively translate procedures into this state machine, enhancing reasoning capabilities. Though not a low-code solution, it offers a schema layer for non-coders to adjust agents. With structured prompts and a memory system comprising long-term (SQL), short-term (dynamic state machines), and transition logs, users can extract valuable insights. Our comprehensive devtools include debugging capabilities, code generators, and more. Following 18 months of development, AI-gent Workflows is now operational, displaying promising performance metrics. Enjoy exploring!
Source link