What TEN is trying to solve, how the compiler is put together, and the milestones between here and a stable 1.0.
Each of these is a deliberate trade-off, not a checklist of buzzwords. Items marked Future are on the roadmap but not shipped.
Native keywords in Tamil, Tanglish, English, with more Indian languages on the roadmap.
Predictable, unambiguous grammar designed for accurate generation and static analysis by AI models.
Compiles through LLVM straight to machine code — no interpreter, no VM overhead.
Types are checked at compile time with clear, actionable diagnostics.
One toolchain targeting Windows, Linux, macOS, and beyond.
Built on a battle-tested optimizing backend used by Rust, Swift, and Clang.
Immutable-by-default design with a borrow checker planned for v1.0.
Learn programming concepts in the language you already think in.
Designed from day one for predictable performance and long-term maintainability.
Keywords are chosen to be easy to speak and dictate aloud, for coding by voice in Tamil, Tanglish, or English. Future
Compiler diagnostics read like plain Tamil (or your chosen language), not cryptic English stack traces.
Convert TEN source between language packs — write in Tamil, generate the Hindi or English version. Future
| Feature | TEN | Python | Rust | Go |
|---|---|---|---|---|
| Multilingual syntax | Yes | No | No | No |
| AI-native grammar | Yes | Partial | No | No |
| Native compilation | Yes | No | Yes | Yes |
| Static typing | Yes | No | Yes | Yes |
| Memory safety | Yes | Partial | Yes | Partial |
| Cross platform | Yes | Yes | Yes | Yes |
Select any stage to see what it does. Source flows top to bottom — language-agnostic from the lexer onward.
TEN's grammar is unambiguous by construction, so AI systems can read, generate, and optimize TEN code with far higher accuracy than languages designed for human convenience alone.
Clear token boundaries and minimal ambiguity make parsing reliable for models.
No hidden context-sensitivity — what you see is what gets parsed.
Tooling can inspect and transform the AST directly, built for codegen pipelines.
Type information stays close to syntax, easing static reasoning.
On-device code assistance is on the long-term roadmap. Future
Every dialect — Tamil, Tanglish, English — compiles to one shared AST.
The canonical AST lets AI models re-render any TEN file into another language pack with no logic drift. Future
Errors, warnings, and hints are generated in the dialect you're coding in, not translated after the fact.
Variables don't change unless explicitly marked mutable.
Errors caught before your program ever runs.
Clear, specific error messages — not stack traces.
Rust-inspired ownership model targeted for v1.0.
Data-race-free concurrency primitives on the roadmap.
No silent type coercion, no surprising runtime behavior.
TEN is pre-alpha. Real benchmarks will be published once they exist — no projected or simulated numbers.
Benchmarked against Python, Java, Go, and Rust once the LLVM backend lands.
Awaiting compiler completion before measurement.
Will be tracked across release binaries.