Show HN: AgentML – Deterministic AI Agents (MIT, Alpha)

    <agentml xmlns="github.com/agentflare-ai/agentml" 
             datamodel="ecmascript">
      <state id="respond">
        <openai:generate model="grok-4-fast-reasoning"
          promptexpr="`continue: ${conversationHistory(10)}`"
          location="_event"/>
        <transition event="done" target="send"/>
      </state>
      <state id="send">
        <send event="output" data="_event.data"/>
        <transition target="respond"/>
      </state>
    </agentml>

AgentML is an open-source XML language for building deterministic AI agents. Write once, run anywhere.

The problem: LLM agents are flaky, locked to specific frameworks, and nearly impossible to debug or audit.

The fix: Declare agent behavior in XML using state machines. State transitions are explicit, outputs are schema-bound, execution is traceable.

Key features:

* No hallucinated tool calls (structured outputs only)

* Built-in memory (SQLite + graph storage)

* 80% fewer tokens via runtime snapshots

* CLI: amlx validate, amlx run

* Swap models freely (OpenAI, Grok, Ollama)

Install: curl -fsSL sh.agentml.dev | sh

Run: amlx run chat.aml

Runtime: Go/WASM (agentmlx). Coming soon: LangGraph export, Python SDK.

GitHub: https://github.com/agentflare-ai/agentml

Docs + Demo: https://www.agentml.dev/

What's your biggest agent pain point - framework lock-in, debugging, or compliance?

Write once, run anywhere.

oh wow! Loving what I am seeing here <3