Skip to main content
Version: latest

X07 (x07lang)

X07 is an agent-first systems language: it’s designed so autonomous coding agents can generate, modify, test, and repair programs reliably—without needing a human to “massage” code.

Most languages optimize for human ergonomics (expressiveness, many equivalent ways to write the same thing). X07 optimizes for:

  • Deterministic behavior (especially in test worlds), so agents can trust feedback.
  • Canonical representations (bytes encodings, module layouts, request formats), so “the same intent” produces “the same shape”.
  • LLM-oriented diagnostics (structured, code-stable error IDs + machine-applicable fixes).
  • Separation of concerns by world: deterministic fixtures for testing vs OS-backed worlds for real work.

The mental model​

Think of X07 as two layers:

  1. A small, stable core (compiler + runtime substrate)
  2. A growing ecosystem of libraries (stdlib + external packages)

Programs are stored and exchanged in a structured AST format (x07AST JSON). Humans usually edit a pretty form (or generated templates), while agents operate on the structured form directly.

Start here​

What makes X07 different?​

1) One canonical way (agents don’t get “choice paralysis”)​

Instead of 5 equivalent ways to read a file, split strings, build output, or handle errors, X07 aims for:

  • one canonical API surface per capability,
  • one canonical bytes encoding per data interchange,
  • one canonical failure model per module (stable error code space).

This reduces “LLM confusion” and makes programs and patches more uniform.

2) Deterministic-by-default testing worlds​

X07’s fixture worlds allow repeatable tests: filesystem trees, request/response cassettes, key/value stores, virtual time, and resource budgets are stable across runs.

That’s how you get:

  • trustworthy regressions,
  • deterministic replay,
  • cheap automated repair loops.

3) Production worlds are opt-in​

When you need real OS resources (real network, real disk, real time), you switch to OS-backed worlds. Those worlds are never used for deterministic evaluation, and are governed by explicit policies.

Documentation map (human)​