Zerolang is a programming language designed for agents that treats the semantic graph as the program database rather than source text.
The language addresses the problem of agent coding loops that treat text as the source of truth, which can lead to agents guessing from source rather than understanding compiler semantics. Zerolang inverts this relationship by making the graph the checked compiler input. Agents interact with explicit semantic handles—symbols, node IDs, graph hashes, types, effects, and ownership facts—rather than manipulating line ranges in text files. The workflow centers on human requests, with agents querying the graph and submitting checked edits that are validated before the store is written. Humans can review and occasionally edit text projections of the graph, but agents primarily use compiler commands like `zero query` and `zero patch` to interact with the underlying semantic structure. This approach prevents stale references, type errors, and invalid shapes from being written to the program database.
Zerolang suits teams building agent-driven systems where safety and semantic correctness matter more than rapid text-based iteration. The tool is explicitly experimental and carries a safety warning about breaking changes and rough edges, making it unsuitable for production systems or sensitive data without isolation. The language prioritizes token efficiency, low memory usage, fast startup and builds, low runtime latency, and small dependency-free artifacts. The projection boundary between graph and text is made explicit through `zero import` and `zero export` commands, preventing human text edits from silently diverging from the graph.
The project maintains an active compiler implementation with commands for querying, inspecting, checking, testing, and running code. Development includes tooling for local iteration and a documented bootstrap skill for agents. The documentation is centralized at the project's homepage, suggesting ongoing effort to guide users through the graph-first model.
