Deterministic, governed runtime stack for auditable computing.
T81 is a deterministic compilation and execution pipeline (T81Lang -> TISC -> HanoiVM). It prioritizes auditability, policy enforcement (Axion), and reproducibility over raw hardware speed.
Note on Floating Point Determinism:
T81Floattranscendental functions (sin,cos,tan,log,exp,sqrt) are now implemented via a deterministic software-defined backend (dmath) and are guaranteed bit-exact across platforms.T81Floatdivision and inverse/hyperbolic trigonometric functions (asin,sinh, etc.) may still rely on host-platform behavior in non-strict modes. Strict bit-exact determinism is guaranteed forT81Int,T81BigInt,T81Fraction(canonical), and coreT81Floatoperations.
Verify the claims yourself in 4 steps:
-
Build & Run Hello World
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build --parallel ./build/t81 compile examples/hello_world.t81 -o hello.tisc ./build/t81 run hello.tisc
-
Run Determinism Gate
# Verify cross-architecture reproducibility hash python3 scripts/ci/t81lang_repro_gate.py --t81-bin build/t81 --check -
Run a VM Demo
./build/t81_demo
-
Inspect a Trace Artifact
./build/t81 trace show trace.txt
To save your time, here is what T81 is NOT:
- NOT a hardware accelerator: T81 does not claim ternary hardware speedups. This is a software runtime for deterministic correctness.
- NOT a general-purpose replacement: T81 focuses on high-stakes, auditable logic, not replacing C++ or Python for general tasks.
- NOT "fast and loose": If a performance optimization breaks trace determinism, T81 rejects it.
Modern runtimes trade reproducibility for speed. T81 inverts this: Auditability is the primary constraint.
This is enforced via a strict architectural boundary between the Language/Compiler and the Execution Runtime, governed by explicit contracts.
View Architectural Boundary Diagram | View Runtime Contract
| Document | Purpose | Authority Scope |
|---|---|---|
| STATUS.md | What is true today | Operational Truth |
| ROADMAP.md | Forward plan | Strategic |
| VERSIONING.md | Compatibility rules | Normative |
| spec/ | Behavioral definition | Normative |
| docs/EVIDENCE.md | Proof of claims | Verification |
- Stable: T81Lang Syntax, TISC Binary Format, HanoiVM Execution Semantics.
- Experimental: JIT Compilation, Distributed Tensor Ops.
- SemVer: T81 follows Semantic Versioning. Breaking changes to Stable components increment the Major version.
| Platform | Compiler | Status |
|---|---|---|
| Linux (x86_64) | Clang 18+, GCC 14+ | ✅ Determinism Gate |
| Linux (ARM64) | Clang 18+ | ✅ Determinism Gate |
| macOS (ARM64) | Apple Clang | ✅ Supported |
git clone https://github.com/t81dev/t81-foundation.git
cd t81-foundation
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failureSingle-threaded safe mode:
cmake --build build --parallel 1
ctest --test-dir build --output-on-failure -j1Common workflows:
# Compile / run
t81 compile examples/hello_world.t81 -o build/hello.tisc
t81 run build/hello.tisc
# Inspect / debug
t81 disasm build/hello.tisc
t81 debug build/hello.tisc
# Diagnostics / reproducibility
t81 check examples/hello_world.t81
t81 repro-hash tests/fixtures/t81lang_determinism
# Trace workflows
t81 trace show trace.txt
t81 trace diff trace_a.txt trace_b.txt
t81 trace replay build/hello.tisc trace.txtModel tooling:
t81 weights import model.safetensors -o model.t81w
t81 weights info model.t81w
t81 weights quantize model.safetensors --to-gguf model.ggufSee full command help:
t81 helpinclude/t81/: public API headerssrc/: frontend, TISC, VM, Axion, CanonFS, CLI implementationtests/: conformance, determinism, VM/e2e, property slicesdocs/: guides, status, benchmarks, runtime boundary docsspec/: normative semantics and governance inputsexamples/: runnable samples and demos
T81 uses an explicit runtime boundary contract:
- Marker:
contracts/runtime-contract.json - Boundary policy:
docs/runtime-semantics-boundary.md
ARCHITECTURE.mddocs/system-integration.mdANALYSIS.mdCHANGELOG.mddocs/research-guide.mddocs/ai-quickstart.md
This repository is licensed under MIT (see LICENSE).