A programming language for AI agents to write, read and maintain.
Statically typed and effect-tracked. Contracts are checked at run time, fuzzed and proved, and it compiles to native code that runs faster than idiomatic C++ with the same safety checks.
type Item = {name: Str, price: Int where _ >= 0, qty: Int where _ > 0} fn total(items: List[Item]) -> Int post r >= 0= items.map(_.price * _.qty).sum fn discount(amount: Int where _ >= 0, pct: Int) -> Int pre pct >= 0 and pct <= 100 post r <= amount= amount - amount * pct / 100 fn main() -> Unit ! log= do pen = Item{name: "pen", price: 3, qty: 4} book = Item{name: "book", price: 12, qty: 1} log("total: {discount(total([pen, book]), 10)}") test empty = total([]) == 0test sale = discount(24, 10) == 22$ sspur run cart.ssptotal: 22$ sspur test cart.ssp2 passed, 0 failedmacOS, Linux and Windows, on arm64 and x86_64. Native compilation uses clang.
brew install sspurlang/sspur/sspur curl -fsSL https://github.com/sspurlang/sspur/releases/latest/download/install.sh | sh It verifies the release's SHA-256 checksum and installs sspur into ~/.local/bin.
irm https://github.com/sspurlang/sspur/releases/latest/download/install.ps1 | iex It verifies the checksum, installs sspur.exe and adds it to your user PATH. Native compilation needs LLVM and the Visual Studio Build Tools.
Other ways to installRelease v1.0.0 and its checksumsEditor setup
0.14xthe time of idiomatic C++ on records, lists and trees, and below 1x on all seven benchmark workloads
An optimizing native compiler, with C++ at clang -O2 and the same safety checks on both sides. Overflow, out-of-bounds access and contract violations always trap; the compiler proves most of those checks away, reuses memory in place and fuses pipelines into single loops.
Every side effect is in the signature, so a reviewer can see what a function may do:
fn save(u: User) -> Unit ! db.write, logpre, post and refinement types like Int where _ > 0 are checked at run time, turned into property tests by the fuzzer and proved with Z3 where possible.
Typed rows with db.query[User], json.decode[T] for any type, and sspur debug on lldb or gdb with SSPUR names.
0.67xthe tokens of Python on 17 multi-step coding tasks, every hidden test passing
Atomic edits that typecheck the whole codebase, typed holes that fill themselves, an MCP server, a language server and editor support from VS Code to Zed.
23packages in the registry, each checked when it was published
sspur add NAME pins a version to a commit and a source snapshot, and the lock pins each dependency by the hash of its exports.
Offset, page-number and keyset pagination: clamped requests, signed cursors, page ranges for UIs, Link headers and JSON envelopes
Liveness, readiness and startup endpoints with concurrent checks (SQLite, HTTP dependencies, files) in the application/health+json format
Number theory, exact rounding modes, primes, combinatorics and statistics for Int, F64 and BigInt, pure and tested in both tiers
Retries with exponential backoff and jitter, retry budgets, a circuit breaker and rate limiters, on a clock effect that tests run virtually
Map utilities (merge with resolvers, invert, filter and map keys and values, counters, multimaps, diffs) and dynamic JSON values with paths, pointers and merge patches
SQLite helpers over db: a parameters-only query builder, keyset pagination, upserts, batched inserts, up/down migrations with checksums, test fixtures