TORQ

The first programming language designed for AI to write.

TORQ is a token-optimized, compiled, general-purpose programming language. AI writes it. Humans approve it. Machines run it fast.

Why TORQ?

Every programming language today was designed for humans to write. TORQ inverts this — it's optimized for AI authorship while remaining readable for human review.

Problem	TORQ's Answer
AI wastes tokens on boilerplate	Zero imports, zero type declarations, sigil-based types
AI picks wrong syntax patterns	One way to do each thing — zero ambiguity
Runtime performance	Compiles to native code, arena-scoped memory, zero GC
Concurrency is hard	Parallel by default, sequential opt-out
Data format juggling	Native JSON, XML, YAML, CSV, TOML parsing built in
Package management	External APIs resolve from OpenAPI/Swagger specs automatically

What It Looks Like

::checkout
  $req.body | as data
  | api.stripe.charge .amount .user
  | retry 3 delay 1s
  | fail -> log.err | respond 500 "payment failed"
  | log "charge complete" | respond 200

::main
  http.listen 8080
  | route post "/checkout" -> ::checkout

That entire checkout service is ~30 tokens for an AI to generate. The Python equivalent is 150+.

Key Features

• Token-minimal syntax — No commas, no semicolons, no imports, no type keywords
• Sigil type system — $ scalar, @ array, % dict, ! error, ~ regex, * shared
• Pipe-based data flow — $data | transform | output
• Parallel by default — Everything concurrent, use sequential to opt out
• Arena-scoped memory — No garbage collector, no borrow checker, zero overhead
• Native data parsing — $raw | as data handles JSON, XML, YAML, CSV, TOML
• 170 built-in functions — Filesystem, HTTP, database, crypto, scheduling — no packages needed
• Two comment types — # prompts (human directs AI, AI deletes after acting) and ## docs (permanent)
• Block-level compilation — Change one block, sub-second rebuild

Quick Start

## hello.torq
::main
  print "hello world"

torq run hello.torq

Project Structure

project/
  torq.yaml          # config, external services, secrets
  src/
    main.torq        # entry point
    *.torq           # organize however you want
  torq_modules/      # auto-generated (gitignored)

Documentation

• Language Specification — Complete language reference
• AI System Prompt — Paste into any AI to enable TORQ generation
• Examples — Working TORQ programs

Bootstrap Strategy

TORQ works with current AI models today — no retraining needed. The grammar is ~30 rules. Paste the AI System Prompt into any model's context and it can write TORQ immediately.

Status

TORQ is in early design and specification phase. The language spec is defined. The compiler is next.

Contributing

TORQ is open source under the MIT License. Contributions welcome.

License

MIT