Flat tool lists are dead. Stateful tools are next.
AgentMLX doesn’t hand an LLM a flat list of APIs — it gives it a state-scoped toolset that evolves with the agent’s behavior.
Memory belongs inside the agent
In AgentML, memory is a first class citizen that lives directly in the runtime.
We’re combining github.com/agentflare-ai/sql… our open sourced openCypher compliant SQLite graph extension, with sqlite-vec to give every agent local and secure memory built in.
From key value to vector search to full graph traversal, memory in AgentML is native, composable, and runs anywhere SQLite runs.
Building AI agents shouldn't require wrestling with UI builders or writing everything from scratch, and I suspect many of you feel the same.
This weekend, I built the first proof of concept AgentML agent designed to create other AgentML agents.
Nearing the completion of our @langchain langgraph transformer.
Watch us build in public as we rapidly work towards a one definition, infinite agent framework paradigm.
We are opening up AgentML a Deterministic Language for Agent Behavior. AgentML combines the usability of XML with the predictability of state machines, giving you complete control over agent behavior. MIT License.
Link in comments.