The systems language built for coding agents, and readable by humans.
Meaning in. Verified machine code out.
Source remains a clear human projection. Coding agents work against typed meaning, stable identity, bounded context, and revision-bound changes instead of rewriting text by guesswork.
revision sha256:<program-state>
query app.main --depth 1
context typed · bounded · stable-id
patch expected_revision == current
verify fail_closed
target native | browser/wasmWhat is Semaprax?
Semaprax is an experimental agent-native systems programming language built around a stable semantic program graph. Its v0.2 prototype lets coding agents query typed meaning, submit revision-bound semantic patches, and lower a verified subset to native executables and browser/Wasm packages.
One program, three working surfaces
Human source, agent-facing meaning, and machine output are derived from one versioned semantic state. Every arrow below is an evidence boundary, not a promise of universal language support.
- Human sourceReadable syntax remains the review and authorship surface.
- Semantic program graphStable IDs expose typed meaning, context, diagnostics, and patch targets.
- Verified outputAdmitted subsets lower deterministically to native or browser/Wasm artifacts.
Semaprax semantic program graph: Semaprax parses human-readable source into validated HIR and a versioned semantic graph. Agents request bounded context by stable ID, preview or submit a patch against an exact revision, and receive deterministic diagnostics or evidence before native or Wasm lowering.
The hypothesis, not the headline
Semaprax tests whether a programming system can reduce the context and repair ambiguity faced by coding agents. Lower model cost is a hypothesis. It is not yet supported by model-token, latency, answer-quality, or repository-scale evidence.
A research index with one owner per question
Each page answers one branded question so search engines and answer engines do not have to guess which route is authoritative.
Architecture
Stable IDs, graph context, and semantic patches
evidenceEvidence
What v0.2 implements and where it stops
benchmarksBenchmarks
Context measurements without cost promises
interoperabilityInteroperability
Target and ecosystem evidence matrix
roadmapRoadmap
Milestones, objectives, and open gaps
Capability evidence ledger
| Current status | Evidence state | Scope | Evidence |
|---|---|---|---|
| Stable semantic program graph | Implemented | Versioned Graph v10-v14 JSON for the documented typed subset, with persistent identities and validated HIR shared by native and Wasm lowering. | GitHub repository |
| Bounded agent context | Implemented | Agent Context v1 plus additive v2 call traversal, with deterministic byte, node, depth, and frontier limits. | GitHub repository |
| Revision-bound semantic patches | Implemented | Atomic Patch v1 and bounded Patch v2 operations, with exact revision checks, stale rejection, and selective semantic delta validation. | GitHub repository |
| Semantic review and evidence capsules | Demonstrated | Bounded review, impact, target, patch, and workspace evidence routes with deterministic artifacts and explicit nonclaims. | GitHub repository |
| Native and browser/Wasm lowering | Demonstrated | Documented scalar, control-flow, record, variant, generic, contract, and bounded ownership slices through C11/Clang and real Node/Wasm gates. | GitHub repository |
| Ownership-inspired memory management | Experimental | Explicit own, borrow, and shared boundaries plus move checking and bounded resource cleanup protocols. This is not Rust compatibility or a complete borrow checker. | GitHub repository |
Questions, answered without the hype
Is Semaprax production-ready?
No. Semaprax is Apache-2.0 pre-alpha research. The repository calls itself an executable architectural seed, and the evidence ledger separates bounded implementation from general completion.
Is Semaprax only for AI agents?
No. The source projection is designed to stay readable by humans. Agent-native describes the stable semantic graph, bounded context, diagnostics, and patch interface available to coding agents.
Need this level of evidence-first engineering in your AI product?
See Wavect’s AI engineering work.