AIXP Labs just published AISOP V1.0.0, an open protocol that lets developers define structured AI agent workflows using either Mermaid diagrams or JSON flow objects—whichever fits their workflow better. The project landed on GitHub with reference implementations in Python and JavaScript, both running on stdlib only. No vendor lock-in, no proprietary builders—just plain JSON that you can version control, edit in any text editor, and feed to any AI runtime.

Why Flow Graphs Beat Freeform Prompts

The core problem AISOP tackles is the chaos of managing multi-step AI programs. Natural language prompts are token-heavy and hard to version. Python or YAML workflows require code. Visual builders like Dify lock you into their ecosystem. AISOP takes a different approach: define your program's control flow as either a Mermaid string (which AIXP claims uses roughly 50% fewer tokens than JSON) or a structured JSON object—and mix both in the same program. The protocol supports branching, parallel fork/join operations, sub-tasks via delegation, error routing with dashed edges, retry policies, and even human-in-the-loop confirmation gates through reserved sys.* system calls.

Fourteen Control Flow Patterns Out of the Box

AISOP ships with 14+ control flow patterns covering everything from simple sequential execution to batch iteration with context isolation. The Mermaid syntax uses familiar shapes—rectangles for processes, diamonds for decisions, circles for end nodes—with solid arrows for normal flow and dashed arrows (Mermaid's '-.->') for error routing. On the JSON side, developers get fields like "next" for sequential/parallel forks, "branches" for conditional logic, "error" for handler routing, "wait_for" with a merge_strategy for joins, and "delegate_to" for calling sub-tasks by name.

Reference Implementations in Python and JavaScript

The reference implementations are intentionally minimal. Both flow_runtime.py and flow_runtime.js require no external dependencies—just standard library calls. The test suites include 44 tests each, and example files demonstrate basic flows plus advanced sys.* system call usage including human confirmation gates via the inviolable sys.io.confirm function. AIXP emphasizes that AISOP is AI-agnostic by design: it works with any LLM backend since it's fundamentally just structured data describing program flow.

The Axiom 0 Foundation

Underpinning the technical spec is "Axiom 0": Human Sovereignty and Wellbeing—the philosophy that AI systems exist to serve humanity. Every AISOP implementation must enforce this axiom at highest execution priority, enforced through the immutable sys.io.confirm system call for human confirmation gates. It's a philosophical stance baked into the protocol itself.

Key Takeaways

  • Zero-dependency reference implementations in Python and JavaScript (stdlib only)
  • Mermaid format cuts token usage by ~50% compared to JSON flow objects
  • 14+ control flow patterns including parallel forks, joins, retries, and human-in-the-loop gates
  • Plain JSON output plays nice with git versioning, any editor, and any AI runtime
  • Apache 2.0 licensed—fully open source from AIXP Labs ecosystem (which includes protocols for agent discovery, commerce settlement, and social trust)

The Bottom Line

AISOP is a clean bet if you're building anything beyond single-prompt LLM interactions. The Mermaid-first approach is smart—diagrams are easier to audit and debug than nested JSON configs. Whether this protocol gains traction depends on whether the AI dev community wants standardization or prefers rolling their own workflow management. Worth watching.