Zero

Product Introduction

1. Definition: Zero (or ZeroLang) is an experimental, graph-first programming language and compiler toolchain developed by Vercel. It is a semantic program database where the source of truth is a structured graph of nodes (symbols, calls, types) and edges, not raw text files.
2. Core Value Proposition: Zero exists to fundamentally optimize the software development workflow for AI agents. Its primary goal is to enable AI-driven coding by providing a deterministic, queryable, and patchable program representation that eliminates the guesswork and toolchain overhead of traditional text-based editing, leading to more efficient and reliable agentic coding.

Main Features

1. Semantic Program Graph as Source: The core program is stored as a compiler-owned graph database (zero.graph). This graph contains semantic nodes for declarations, expressions, types, and effects, linked by edges. Agents interact directly with this structured data.
2. Compiler-Checked Graph Patching: Agents modify the program by submitting atomic patch operations (zero patch) to the compiler. Each patch specifies target nodes, expected current values (using graph hashes), and new values. The compiler validates the patch for type safety, graph structure, and state consistency before application, preventing invalid states.
3. Human-Readable Code Projections: The compiler can generate readable text projections (e.g., src/main.0) from the graph at any time for human review, debugging, or rare manual intervention. The projection is a derived artifact, not the primary source.
4. Deterministic Build & Runtime Characteristics: Zero is engineered from the ground up for performance in agentic and serverless environments. It prioritizes token efficiency for AI interactions, fast startup times, low memory footprint, rapid compilation, and has zero external dependencies.

Problems Solved

1. Pain Point: The inefficiency and unreliability of AI agents working with text-based source code. The traditional loop (write text -> run formatter/linter/compiler -> parse errors -> guess intent -> repeat) is slow, token-expensive, and prone to cascading errors.
2. Target Audience: Primarily AI coding agents and their developers (e.g., builders of AI-powered IDEs, agentic workflows, and coding assistants). Secondarily, engineers and researchers exploring the future of programming languages and AI-human collaboration.
3. Use Cases: Automating API construction (e.g., "build a CRM API"), adding standardized features (e.g., "add authentication"), refactoring code, fixing bugs identified by tests, and generating code from high-level human requests within a safe, checked system.

Unique Advantages

1. Differentiation: Unlike traditional languages (Go, Rust, Python) or even newer AI-centric tools that wrap existing compilers, Zero inverts the paradigm. The graph is the program. This contrasts with tools that use AI to generate text for a traditional compiler; Zero uses the compiler to validate edits for an AI.
2. Key Innovation: The integration of a cryptographic graph hash into the patching workflow. This allows the agent and compiler to agree on the exact program state being edited, preventing stale data conflicts and ensuring patches are applied to the intended semantic version, which is a critical requirement for concurrent, non-linear agentic editing.

Frequently Asked Questions (FAQ)

1. What is Zero programming language used for? Zero is an experimental language designed specifically for environments where AI agents are the primary code writers. It is used to research and enable more reliable, efficient, and deterministic AI-driven software development by having agents edit a semantic program graph directly.
2. How does ZeroLang work with AI agents? AI agents interact with Zero by querying the program graph to understand code structure and then submitting verified patch operations to the Zero compiler. The compiler checks each patch for type safety and consistency before updating the graph, creating a tight, validated feedback loop.
3. Is Zero language ready for production? No, Zero is explicitly an experimental project from Vercel. The documentation states to expect breaking changes and to run it only in safe environments, not against production systems. It is currently a research and exploration platform.
4. What are the benefits of a graph-based programming language? A graph-based language like Zero offers benefits for AI: unambiguous semantic structure for queries, atomic and verifiable state changes via patches, inherent prevention of syntax errors, and the ability to manage code as a database, which is more natural for automated reasoning than text.
5. Can humans write Zero code directly? The primary interface for humans is high-level requests and reviewing the generated code projections (.0 files). While the projection is readable and editable, the intended workflow is for humans to work through agents. Direct graph editing is a low-level operation meant for the compiler and agents.