Skip to main content

The One Diagram That Explains OpenExec

To understand OpenExec, you must first understand the fundamental shift in how it handles the AI control plane compared to traditional tools like Claude Code, Cursor, or Codex.

1. How most AI coding tools work​

In traditional tools, the LLM is the runtime. The system behavior depends on prompts, probabilistic retrieval, and hidden orchestration. Developers try to control it using CLAUDE.md or system prompts, but these are mere suggestions, not deterministic logic.

graph TD
    User((User)) --> Tool[AI Coding Tool<br/>Claude Code / Cursor / etc.]
    Tool --> LLMRuntime[LLM Runtime]
    
    subgraph LLMRuntime [LLM Runtime]
        Rules[Prompt Rules<br/>CLAUDE.md]
        Tools[Tool Calls]
        RAG[RAG / Vector DB]
        Orch[Hidden Orchestration]
    end
    
    LLMRuntime --> Repo[Repository + Terminal]
    
    style User fill:#f9f,stroke:#333,stroke-width:2px
    style LLMRuntime fill:#fff3f3,stroke:#ff0000,stroke-width:2px,stroke-dasharray: 5 5

2. OpenExec Architecture​

OpenExec flips the control plane. The runtime is deterministic and governed by code, while the LLM is treated as a reasoning component that helps inside strict boundaries.

graph TD
    User((User)) --> Runtime[OpenExec Runtime<br/>Deterministic]
    
    subgraph Runtime [Deterministic Runtime]
        direction TB
        BE[Blueprint Engine<br/>Workflow Logic]
        PE[Policy Engine<br/>Security + Permissions]
        OM[Operational Memory<br/>Pointer Records]
        CB[Context Builder<br/>Safe Hydration]
    end
    
    Runtime --> LLM[LLM<br/>Claude / GPT / Gemini / Local]
    LLM --> Actions[Typed Tool Actions]
    Actions --> Systems[Workspace / Systems]
    
    style User fill:#f9f,stroke:#333,stroke-width:2px
    style Runtime fill:#f0fff0,stroke:#00aa00,stroke-width:3px
    style Actions fill:#fff,stroke:#333,stroke-width:2px

3. The Difference in One Sentence​

  • Most tools: The LLM decides what happens.
  • OpenExec: The Runtime decides what happens; the LLM helps reason inside enforced boundaries.

4. Why This Matters​

  • Enforcement vs. Suggestions: In a prompt-based system, "always run tests" is a suggestion the model may ignore. In OpenExec, run_tests is a blueprint stage that always executes.
  • Maintenance vs. Writing: Traditional tools are optimized for writing new code. OpenExec is optimized for maintaining real systems because it combines operational memory with deterministic workflows.
  • Model Portability: OpenExec uses a model adapter layer. Switching from Claude to Gemini or a local model does not change your system logic or security policies.

5. The Mental Model​

The easiest analogy for engineers is this:

  • Claude Code : AI coding = Docker : containers
  • OpenExec : runtime orchestration = Kubernetes : containers

OpenExec is the runtime infrastructure for AI-assisted engineering. Once you treat the LLM as a component in a larger system rather than the system itself, the architecture makes perfect sense.