Zero
An experimental graph-first programming language built for AI agents.
Last verified:
What is Zero?
Zero is an experimental programming language where the graph structure is the program, designed for AI agents to collaborate with humans. Agents submit checked semantic edits to the program graph, with a compiler validating patches before application to prevent stale-state or invalid changes. It aims to create tighter agent loops by putting the compiler in the iteration cycle rather than relying on text-based source code.
Zero pricing
Pricing model: Freemium
Zero pros
- Graph-based semantics enable agents to edit program structure directly with compiler validation, preventing invalid changes
- Checked by default: graph patches are guarded by hashes and expected values, failing safely before touching the store
- Readable projections remain available for human review and manual edits alongside agent-driven changes
- Runtime-optimized design targets small binary size, low memory, fast builds, and zero dependencies
Zero cons
- Explicitly experimental with breaking changes expected; unsafe for production systems
- Early-stage technology with limited ecosystem and documentation
- Requires agents to understand graph structure and patch semantics, adding complexity to agent implementation
Frequently asked questions about Zero
How is Zero different from traditional programming languages?
Zero uses a graph as the program database instead of text. Agents submit checked patches to semantic nodes; the compiler validates changes before application, eliminating the traditional text→format→check→build→test cycle.
Can I use Zero in production?
No. The page explicitly warns: 'Expect breaking changes. Run it in a safe environment, not against production systems.'
How do agents interact with Zero?
Agents query the program graph (symbols, calls, types, effects, node IDs), submit patches with expected graph hashes and values, and the compiler validates before application. Humans then review readable projections.
What are Zero's design goals?
Stay small, fast, explicit, and dependency-free while reducing guessing in the agent loop through compiler-checked semantic edits.