About VVM

You've been programming a computer without realizing it.

Every time you use Claude Code, Cursor, or Codex, you're instructing a machine that can read files, write code, execute commands, and iterate on its own outputs. That's not an assistant. That's a general-purpose computer that understands meaning.

We've been programming it with chat messages. That works until it doesn't.

When English Breaks Down

Simple tasks are fine. "Refactor this function" needs no specification.

Complex tasks fall apart. You want three analyses to run in parallel, feed into a synthesis, retry on failure, and only proceed if the output meets a quality bar. You can say that in English. But which parts are instructions and which are suggestions? What happens if you're ambiguous about the retry logic? The model will do something. It might not be what you meant.

English handles intent. It can't handle structure.

The Model Is the Runtime

VVM is a programming language for AI sessions.

You write a program. You hand it to the model. The model becomes the runtime.

This inverts the usual pattern. Frameworks like LangChain put the orchestration in your code and treat the model as a function to call. VVM puts the orchestration inside the model. The intelligence doesn't just execute steps—it interprets the program, manages dependencies, and makes decisions about how to proceed.

That's not just cleaner architecture. It enables something new.

Predicates That Understand

When orchestration lives in your Python, branching conditions must be things Python can compute. So you write if confidence_score > 0.8—a proxy metric, trying to capture "is this good enough?" in a number.

When orchestration lives in the model, conditions can be semantic. "Is this production ready?" isn't a threshold. It's a question the runtime can answer by reading and judging. The model operates in meaning-space. Now your programs can too.

Open Standard

VVM is open source.

Today that's Claude Code. Codex, Amp, and OpenCode are planned. The language is runtime-agnostic by design.

FAQ