Install Quidra and run a program

01 / install

Download a release

Every release publishes archives for Linux, macOS and Windows on GitHub. Pick the one for your platform, then put the quidra executable on your PATH.

• Clang 15+ for quidra FILE.qui, quidra run and quidra build (native code generation).
• LLVM 15+ with lli for the REPL's ORC JIT.

Quidra 0.5.0 was released on 2026-10-13. Release notes and checksums are on the release page.

sudo apt-get install ./quidra-linux-amd64.deb
quidra --version

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failure
./build/quidra run examples/hello.qui

02 / first program

Write, run, build

Source files end in .qui. Top-level statements are the entry point, blocks are four-space indented, and there is no main to declare.

quidra hello.qui compiles and runs natively, reusing a verified executable from the operating system's user cache when its inputs are unchanged. quidra build keeps the executable.

print("Hello from Quidra")
print(NL)

Hello from Quidra

quidra hello.qui

quidra build hello.qui -o hello
./hello

03 / REPL

Evaluate interactively

Run quidra with no arguments in a terminal. Each submission is parsed, checked, lowered to LLVM IR and executed natively through ORC JIT; declarations stay available for later submissions, and a rejected submission leaves the session intact.

:type expression prints a static type, :reset clears the session, :help lists the rest.

$ quidra
Quidra 0.5.0
>>> int x = 5
>>> x * 8
40
>>> int square(int value)
...     return value * value
...
>>> square(6)
36

04 / tooling

Let the compiler explain itself

The same frontend that compiles your program answers questions about it. Diagnostics have stable codes and exact spans, formatting is canonical, and the typed IR and generated LLVM IR are one command away.

A failed check is data, not prose. Here is the JSON for an uninitialized read:

quidra check program.qui --json    # diagnostics with codes and exact spans
quidra fmt program.qui --check      # verify canonical formatting
quidra ir program.qui               # typed Quidra IR
quidra llvm program.qui             # generated LLVM IR
quidra inspect program.qui          # typed source nodes with ids and hashes
quidra lsp                          # Language Server Protocol over stdio

int x
print(x)
print(NL)

{"ok":false,"diagnostic_schema_version":2,"truncated":false,"diagnostics":[{"code":"UNINITIALIZED","message":"'x' is read before it is initialized on every path.","file":"uninitialized.qui","source_revision":"b4a7958a3d9dc54b5b236a70f44e72232e9c461d11e3cff75b807cadb32a78e6","node_id":"n000003","node_kind":"name","span":{"start":{"offset":12,"line":2,"column":7},"end":{"offset":13,"line":2,"column":8}}}]}

05 / packages

Install first-party packages

Packages are source packages installed from immutable release tags, never from a moving branch. Official distributions use the quidra- prefix; the import name stays short.

Each package release declares the compiler range it supports in quidra.package; with no version written, the newest release compatible with your compiler is chosen, and an incompatible request fails instead of silently picking another version. The synchronized v0.5.0 releases of Math, NN, Vision, Video and DNN support Quidra 0.5.0. See package management for the full contract.

quidra install quidra-vision   # newest release compatible with this compiler
quidra install quidra-dnn
quidra list
quidra lock program.qui         # pin versions and SHA-256 in quidra.lock

import vision
import dnn

06 / in the browser

Or start in the browser

The Quidra Playground runs the real compiler frontend as WebAssembly for browser-local Check, Format, IR, Inspect and Patch. Build and Run send source only when you request them to a separately sandboxed native runner, and stay disabled unless its language version and exact Core commit match the WebAssembly frontend.

Open the Playground

int count = 3
auto name = "Quidra"
print("Hello, {name}")
print(NL)

int[] original = [1, 2, 3]
int[] copy = original
copy[0] = 9
print("original[0] = {original[0]}, copy[0] = {copy[0]}")
print(NL)

int total = 0
for auto value = original[]
    total += value
print("total = {total}")
print(NL)

for 0 <= int i < count
    if i == 0
        print("first")
        print(NL)
    elif i < count - 1
        print("middle")
        print(NL)
    else
        print("last")
        print(NL)

Hello, Quidra
original[0] = 1, copy[0] = 9
total = 6
first
middle
last