Why Objeck?
Built for modern development:
- 🚀 JIT-compiled for performance (ARM64/AMD64)
- 🤖 AI-native: OpenAI, Gemini, Ollama, ONNX, OpenCV — no third-party packages
- 🌐 Network-complete: HTTP/1.1 · HTTP/2 · HTTP/3/QUIC · WebSocket · DTLS — all standard library
- 💻 Developer-friendly: REPL shell, LSP plugins for VSCode/Sublime/Kate, DAP debugger
- 🌍 Cross-platform: Linux, macOS, Windows (x64 + ARM64, including 64-bit Raspberry Pi)
- 🔧 Full-featured: Threads, generics, closures, reflection, serialization
Perfect for:
AI/ML prototyping • Computer vision • Web services • Real-time applications • Game development
Try It Online
👉🏽 Playground — 42 demos across 10 categories, Monaco editor, no install required.
Quick Start
# Ubuntu 24.04: one package, system libraries included (-arm64 for ARM)
curl -LO https://github.com/objeck/objeck-lang/releases/download/v2026.10.1/objeck-ubuntu24.04-x64_2026.10.1.deb
sudo apt install ./objeck-ubuntu24.04-x64_2026.10.1.deb
# Any other Linux, or macOS: the archive
curl -LO https://github.com/objeck/objeck-lang/releases/download/v2026.10.1/objeck-linux-x64_2026.10.1.tgz
tar xzf objeck-linux-x64_2026.10.1.tgz
# Linux only: install the system libraries the toolchain links against
# (mbedTLS, readline, SDL2/GL, OpenCV, unixODBC, LAME) --
# obr does not start without them. --check reports without installing.
./objeck-lang/install_deps.sh
# Sets PATH and OBJECK_LIB_PATH for this shell. Source it, do not run it.
source ./objeck-lang/env.sh
# Hello World
echo 'class Hello {
function : Main(args : String[]) ~ Nil {
"Hello World"->PrintLine();
}
}' > hello.obs
# Compile and run (modern syntax)
obc hello && obr hello
📖 Full docs: objeck.org
💡 Examples: github.com/objeck/objeck-lang/programs
What's New
Full details for every release are in the CHANGELOG.
v2026.10.2
- Query.Structured — a columnar database in the standard library: SQL with joins, grouping, indexes and expressions; files, URLs, archive members and worksheets named in
FROM; a crash-safe .otab file format; sources that read CSV, JSON, XML, text logs, filesystems, zip/tar, Excel, ODBC, ML results and image detections as tables (design)
- ML and NLP lab — kernel and one-class SVMs, outlier detectors and an autoencoder,
System.ML.Lab for experiments, result tables and charts, and BERT sentence embeddings over ONNX
- Faster runtime — a size-class allocator, a collection threshold that follows the live heap, an open-addressed
Hash, an O(1) Cache, and unboxed IntVector/IntMap/BitSet
- Calls nest 16,384 deep (
--call-depth), and a full stack is recoverable compiled or interpreted
- Networking — DTLS works for the first time, servers take IPv4 and IPv6 clients on every platform or bind one address, every socket failure says why, blocking I/O no longer stalls the collector, UTF-8 URLs, every HTTP status code
- Non-ASCII text on Windows — arguments, environment variables, paths, install directories and console output, in every tool
- JIT fixes —
Float->NaN()/Inf() and two-dimensional array code are compiled instead of handed back to the interpreter, and a value is no longer lost at a loop's collection safepoint
- Garbage collector fix — a conditional's value (
c ? a : b) survives another thread's collection; multithreaded programs could store a dangling reference
- Language server about 2× faster per edit — each edit is analyzed once instead of twice
- 32 new examples — numbered in learning order, each with its compile and run lines and the output it prints, written for people new to the language
- Compiler and library fixes — a quote inside
{$...} is an error instead of a compiler hang; a bounded generic class returns its type argument and a literal in {$...} compiles; ambiguous names, numeric literals (0o is octal; 010 is an error), array literals; CSV, JSON (escapes are decoded on parse), regex, NLP, ML, Ollama, OpenAI, Gemini, SDL2 and ODBC, including a macOS driver manager that could load no driver
v2026.10.1 ✅
- Typed errors —
Runtime->Raise, and try/on/finally that catch by error type
- Named arguments, method-level generics and
ref parameters
- Array literals hold expressions; jagged arrays
- Language server rebuilt around LSP 3.17
- Debian packages for Ubuntu 24.04, plus fixes to the MSYS2 build, OpenCV 5 support and release verification
v2026.10.0
Try() and ?-> recover reliably inside JIT-compiled code- Any ONNX model can be run, and
API.Inference reaches model servers over HTTP
- Signed release manifests that
obu verifies
- ML evaluation: thresholds, bootstrap confidence intervals and per-group recall
Downloads
Latest Release: v2026.10.1
📦 Alternative: Sourceforge • 📚 API Docs: objeck.org/api/latest
Note: Windows installers are signed and timestamped (CN=Randy Hollines, Sectigo); the macOS .pkg is signed and notarized. Signing uses a hardware token and therefore happens locally after publication, so verify rather than assume — Get-AuthenticodeSignature <file>.msi reports Valid only when it really is signed. Check any download against the release's SHA256SUMS, which is regenerated after signing. Builds are automated on GitHub Actions runners.
Verifying a download
Every release ships SHA256SUMS and an Ed25519 signature over it,
SHA256SUMS.minisig. Download both beside the archive:
obu verify objeck-linux-x64_<VERSION>.tgz SHA256SUMS
obu checks the signature before it reads a line of the manifest, and refuses a
manifest it cannot verify. Exit codes are distinct on purpose: 0 matches, 1
does not match, 2 could not be checked.
The manifest matters because it is served from the same place as the assets it
describes — anyone able to replace an archive could replace the manifest to match.
The Linux and macOS archives carry no platform signature at all, so for those it
is the only integrity claim there is. The signature is what an attacker cannot
reproduce.
The release public key, for checking out of band — with minisign, or against
what obu reports:
untrusted comment: minisign public key 8237C6B17C7EA6B9
RWS5pn58scY3gv7X3EL/fJaOQli1V0HDwwFVt+2IWqPXBxvE4U1UhxHC
Key ID 8237C6B17C7EA6B9. The same key is committed at
core/release/objeck-release.pub and served,
independently of this repository, from
https://www.objeck.org/objeck-release.pub. A key you can only obtain from
the thing you are trying to verify proves nothing, which is why it is in more
than one place. Check that they agree:
curl -sS https://www.objeck.org/objeck-release.pub | sha256sum
# 864c6a9951a539ec356520dbeca4f97ec7a5fc9f4fd31d881f659f59bf7a6c04
That digest is of the whole 113-byte file, both lines, newline included. The
site serves .pub as application/x-mspublisher, so a browser offers to
download it rather than showing it; for reading in a browser there is an
identical copy at
objeck-release.pub.txt,
served as text and with the same digest. Design and
key-rotation procedure: docs/release_integrity.md.
See It In Action
HTTP/2 Client
use Web.HTTP;
# Persistent connection — multiple requests share one TLS session
client := Http2Client->New("httpbin.org");
resp := client->Get("/get");
"Status: {$resp->GetCode()}"->PrintLine(); # Status: 200
body := "{\"lang\":\"objeck\"}"->ToByteArray();
resp2 := client->Post("/post", body, "application/json");
client->Close();
# One-liner for quick requests
resp := Http2Client->QuickGet(Url->New("https://httpbin.org/get"));
HTTP/3 / QUIC Client
use Web.HTTP;
# QUIC over UDP — zero round-trip connection on repeat visits
client := Http3Client->New("quic.nginx.org");
resp := client->Get("/");
"Status: {$resp->GetCode()}"->PrintLine(); # Status: 200
client->Close();
# One-liner
resp := Http3Client->QuickGet(Url->New("https://quic.nginx.org/"));
AI Integration
# OpenAI Realtime API - get text AND audio
response := Realtime->Respond("How many James Bond movies?",
"gpt-4o-realtime-preview", token);
text := response->GetFirst();
audio := response->GetSecond();
Mixer->PlayPcm(audio->Get(), 22050, AudioFormat->SDL_AUDIO_S16LSB, 1);
Face Recognition
# SCRFD detector + ArcFace R50 embeddings (InsightFace buffalo_l)
session := FaceSession->New("det_10g.onnx", "w600k_r50.onnx");
r1 := session->Recognize(img1_bytes, 0.5);
r2 := session->Recognize(img2_bytes, 0.5);
faces1 := r1->GetResults(); faces2 := r2->GetResults();
sim := FaceSession->Compare(faces1[0]->GetEmbedding(), faces2[0]->GetEmbedding());
"Same person: {$(sim > 0.35)}"->PrintLine();
Computer Vision
# OpenCV: grayscale, blur and edge detection
image := Image->Load("photo.jpg");
gray := image->CvtColor(ColorConversionCodes->BGR2GRAY);
edges := gray->GaussianBlur(5, 5, 0.0)->Canny(50, 150);
edges->Save("edges.jpg");
Natural Language Processing
# Sentiment analysis and TF-IDF
text := "This product is absolutely wonderful!";
sentiment := SentimentAnalyzer->Classify(text); # "positive"
# Train TF-IDF on documents
docs := ["cats are pets", "dogs are pets", "birds can fly"];
tfidf := TF_IDF->New();
tfidf->Fit(docs);
vector := tfidf->Transform("cats and dogs"); # [0.47, 0.0, 0.47, ...]
🎯 More examples
Language Features
Object-Oriented
- Inheritance, interfaces, generics
- Type inference and boxing
- Reflection and dependency injection
- See OOP examples →
Functional
Strings & Formatting
- Interpolation with expressions:
"{$i + 1}", "{$obj->M()}"
- Format specifiers:
"{$pi:.2}", "{$n:05}", "{$v:x}"
- Positional templates:
String->Format("{0} = {1}", a, b)
- See string features →
Platform Support
Libraries
AI & Machine Learning — 📖 AI Developer Guide · GitHub source · 🤖 Getting Models
- OpenAI — chat, vision, realtime audio, image generation, embeddings, moderation, batch
- Gemini — chat, vision, search grounding, files, context caching, batch embeddings
- Ollama — local LLM chat, vision, and embeddings; recommended models:
llama3.2, phi3, llava (get models →)
- NLP — tokenization, TF-IDF, text similarity, sentiment analysis
- OpenCV — computer vision: detection, transforms, video
- ONNX Runtime — local ML inference: YOLO, ResNet, DeepLab, OpenPose, Phi-3, face recognition (get models →)
- Face Recognition — SCRFD detector + ArcFace R50 (InsightFace buffalo_l)
- Phi-3 / Phi-3 Vision — local SLM text and multimodal inference
Web & Networking
Data
Graphics & Gaming
Other
Architecture
One compiler front end serves obc and the language server; the VM interprets, then JIT-compiles hot methods for AMD64 and ARM64; libraries reach SDL2, ONNX Runtime, OpenCV, mbedTLS and more through native modules. More diagrams.
Development
Modern tooling and practices:
- 🤖 Claude Code for pair programming, debugging, and refactoring
- 🔄 CI/CD: Fully automated build, test, sign, and release pipeline (GitHub Actions)
- ✅ Every push triggers multi-platform builds (Windows, Linux, macOS)
- ✅ macOS installers signed and notarized in CI (Windows signing is a local step — hardware token)
- ✅ One-tag releases:
git tag v2026.2.1 → automated distribution in 60 minutes
- ✅ Parallel builds across 6 platforms (x64/ARM64)
- 📖 Release Process Documentation • CI/CD Architecture • System Architecture
- 🔍 Quality: CodeQL security scanning + gitleaks secret scanning
- 🧪 Testing: 350+ tests across 3 suites (regression, comprehensive, deploy)
- Regression suite: 10 focused tests for critical functionality
- Comprehensive suite: 323+ tests for full language validation
- Deploy suite: 17 real-world usage examples
- Full cross-platform coverage (Windows/Linux/macOS, x64/ARM64)
Editor Support:
📚 Testing Documentation • 🧪 Regression Tests • 📊 Performance & Benchmarks
Resources
