AgentML – SCXML for Deterministic AI Agents (MIT)
Hey HN, We’ve been experimenting with how to make AI agents more deterministic, observable, and production-safe, and that led us to build AgentML — an open-source language for defining agent behavior as state machines, not prompt chains. My co-founder posted before but linked to the project website instead of the repo, so resharing here. AgentML lets you describe your agent’s reasoning and actions as a finite-state model (think SCXML for agents). Each state, transition, and tool call is explicit and machine-verifiable. That means you can: - Reproduce any decision path deterministically -…
What it does
In the maker’s words, at launch
Hey HN, We’ve been experimenting with how to make AI agents more deterministic, observable, and production-safe, and that led us to build AgentML — an open-source language for defining agent behavior as state machines, not prompt chains. My co-founder posted before but linked to the project website instead of the repo, so resharing here. AgentML lets you describe your agent’s reasoning and actions as a finite-state model (think SCXML for agents). Each state, transition, and tool call is explicit and machine-verifiable. That means you can: - Reproduce any decision path deterministically - Trace reasoning and tool calls for debugging or compliance - Guarantee agents only take valid actions (e.g. “never send a payment before verification”) - Run locally, in the cloud, or within MCP-based frameworks Example: ``` <?xml version="1.0" encoding="UTF-8"?> <agentml xmlns="github.com/agentflare-ai/agentml" xmlns:openai="github.com/agentflare-ai/agentml-go/openai" version="1.0" datamodel="ecmascript" name="researcher"> <datamodel> <data id="papers" expr="[]" schema='{"type":"array","description":"Fetched papers from Hugging Face"}' /> <data id="summary" expr="''" schema='{"type":"string","description":"Summary of the papers"}' /> </datamodel> <state id="start"> <onentry> <log label="Researcher: " expr="`Fetching papers from Hugging Face and summarizing with OpenAI\n`" /> <openai:generate model="gpt-4o" location="summary" stream="false"> <openai:prompt>Summarize these recent AI/ML papers from Hugging Face: {{fetch "https://huggingface.co/api/daily_papers"}} Provide a concise summary of the key trends, breakthroughs, and developments in AI/ML research. </openai:prompt> </openai:generate> </onentry> <transition target="log_summary" /> </state> <state id="log_summary"> <onentry> <log label="Researcher Summary: " expr="summary" /> </onentry> <transition target="done" /> </state> <final id="done" /> </agentml> ``` We’re using this in Agentflare to add observability, cost tracking, and compliance tracing for multi-agent systems — but AgentML itself is fully open-source (MIT licensed). Repo: https://github.com/agentflare-ai/agentml Docs: https://docs.agentml.dev We also launched SQLite-Graph, a Cypher-compatible graph extension for SQLite, which will serve as the base for AgentML’s native memory layer. It’s also MIT licensed: https://github.com/agentflare-ai/sqlite-graph Would love feedback from anyone building with LLM orchestration frameworks, rule-based systems, or embedded MCP tool servers… especially around how to extend deterministic patterns to multi-agent coordination. — Jeff @ Agentflare
Does the same job
all alternatives →
- FAFirst autonomous ML and AI engineering AgentJan 2026 · marketplace.visualstudio.com · ▲5
Founder here. I built NEO, an AI agent designed specifically for AI and ML engineering workflows, after repeatedly hitting the same wall with existing tools: they work for short, linear tasks, but fall apart once workflows become long-running, stateful, and feedback-driven. In real ML work, you don’t just generate code and move on. You explore data, train models, evaluate results, adjust assumptions, rerun experiments, compare metrics, generate artifacts, and iterate; often over hours or days. Most modern coding agents already go beyond single prompts. They can plan steps, write files, run…
- ACAgent-contracts, contract-based LangGraph agentsJan 2026 · github.com · ▲8
Hi HN, I’m the author of agent-contracts, a Python library that explores a contract-based approach to structuring LangGraph agents. When building larger LangGraph-based systems, I kept running into the same issues: - node responsibilities becoming implicit - state dependencies spreading across the graph - routing logic getting harder to reason about - refactoring feeling increasingly risky agent-contracts is an attempt to make these boundaries explicit. Each node declares a contract that describes: - which parts of the state it reads and writes - what external services it depends on - when…
Maritime, a platform for running AI agents for $1 a month18d ago · maritime.sh · ▲8Hi HN, my name is Maria, and I’m a co-founder of Maritime. We started Maritime at MIT to build infrastructure for companies that need to run thousands of isolated AI agents for their customers. Imagine you set up an agent like OpenClaw, or a personal assistant agent with a custom framework, and want to give a separate version of it to every customer/friend. Each customer needs their own agent running in an isolated microVM, with persistent state, secrets, triggers, and sleep/wake behavior. Building such scalable and secure infra will take you months and will cost hundreds of…
- AAAgentic AI Frameworks on AWS (LangGraph,Strands,CrewAI,Arize,Mem0)2025 · github.com · ▲7
We’ve published a set of open-source reference implementations on how to build production-grade Agentic AI applications on AWS. What’s in the repo: • Agentic RAG, memory, and planning workflows with LangGraph & CrewAI • Strands-based flows with observability using OTEL & Arize • Evaluation with LLM-as-judge and cost/performance regressions • Built with Bedrock, S3, Step Functions, and more GitHub: https://github.com/aws-samples/sample-agentic-frameworks-on-... Would love your thoughts — feedback, issues, and stars welcome!
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, November 2025
the whole month →
- IB
Life & fun · Nov 2025 · bitsnpieces.dev
- BBoing▲782
Life & fun · Nov 2025 · boing.greg.technology