About X07
X07 is built for a world where most code is generated — and where the scarce resource is trust in running it.
That changes what matters:
- Determinism beats “cleverness”
- Canonical encodings beat “flexibility”
- Machine-readable diagnostics beat “nice error strings”
- Evidence (proofs, budgets, replayable runs) beats review-by-reading
- Stable boundaries beat ad-hoc conventions
Some high-level concepts that make those goals practical:
- Streaming pipes (
std.stream.pipe_v1) for deterministic, budgeted streaming composition - Branded bytes (
bytes@B) for typed boundary encodings - Structured concurrency (
task.scope_v1) for “no orphan tasks” - Record/replay (
std.rr) for OS → deterministic cassettes - Budget scopes (
budget.scope_v1) for local cost contracts - Contracts tooling: architecture check, schema derive, state machines
- Property-based testing: PBT (
x07 test --pbt) with repro → regression promotion (x07 fix --from-pbt) - Certificate-first verification: Formal verification & certification (
x07 verify --prove,x07 prove check,x07 trust certify) - Human review artifacts: Review & trust (
x07 review diff,x07 trust report)
On authoring: agents and humans can write X07 directly — the x07text projection, x07 doc behavioral summaries, and did-you-mean diagnostics (all 2026-06) exist for exactly that. Whether direct authoring deserves deeper language investment is decided by a published comparative eval, not by assertion (see Why X07 and the roadmap).
X07 is open-source, with direction guided through:
- clear technical governance,
- strict compatibility policies for pinned specs,
- a small core toolchain and a deliberately narrow set of active companion repos (2026-06 scope).
What X07 is not
- A sandbox for running hostile, untrusted code.
- A language that optimizes primarily for human hand-writing.
- A language that allows unlimited “styles” for the same task.
- A proven win for direct agent authoring — that claim is gated on the comparative eval in
labs/agent-eval/.
What X07 is
- A deterministic, certifiable execution substrate for agent-written software.
- A practical C-backend systems language.
- A library ecosystem designed around pinned contracts.