pre-v1 · 0.0.19 · v1.0 codename Ludens

.pseudo

The language built for humans and AI agents.

A typed pseudocode language that compiles deterministically to 11 native targets. Comments are language protocol. Tools-first execution. Local + remote RAG over the entire corpus.

Join the v1.0 waitlist →Fund Pseudo →Build Pseudo →Read the spec View on GitHub

/// @intent     Hash a password using argon2id with OWASP-exceeding defaults.
/// @security   Side-channel resistant; constant-time comparison.
/// @conforms   RFC 9106
/// @author     claude-opus-4-7  (2026-05-15)
pub fn hash_password(plaintext: string) -> Result<HashedPassword, HashError> {
  let salt = crypto.random_bytes(16)?
  let hash = argon2id::hash(plaintext, salt, OWASP_2026)?
  return Ok(HashedPassword { value: hash })
}

Compiles to TypeScript · Rust · Python · Go · Swift · Kotlin · C# · C++ · WASM · shell · Ruby

Why Pseudo

Five things no other language gives you on day 1

11 native targets, one source

Author once. Ship native binaries on every platform — web, mobile, Apple, Windows, Linux, edge — without a runtime.

Comments are language protocol

Every pub item MUST carry @intent + @author doc comments — enforced by the parser, not a linter. Per-fn @conforms RFC 9106 makes standards-conformance auditable per function.

Multi-agent first-class

Eight AI coding agents author and review Pseudo in parallel branches. Claude Code Opus 4.7 max-effort + 1M context is the mandatory final reviewer on every PR.

Secure by default

argon2id, SSRF-blocked HTTP clients, Tink-style typed crypto keys, Paseto v4 tokens, Helmet headers — opt-out is greppable.

No aspirational tiers

Every entry in COMPATIBILITY.md and CONFORMANCE.md ships fully working / fully conforming on v1.0 day 1. No "best-effort," no "should conform," no "v2 target." If it's listed, it works. If it's not ready, it's in the explicit out-of-scope section — not soft-listed.

Compatibility · committed v1.0 day 1

11 emit targets

One source, eleven native targets (per ADR-0019 + ADR-0032 + ADR-0039). Each is a v1.0 commitment, not a goal.

Target	Min version	Notes
TypeScript	5.8 + ES2024	Node 22+ · Bun 1.3+ · Deno 2 · Temporal default in Node 26
Rust	1.88 + Edition 2024	Tokio LTS 1.51 · jiff date/time · async closures stable
Python	3.13 (3.14 opt-in)	PEP 695 type syntax · PEP 779 free-threading supported in 3.14+
Go	1.26	Range-over-func iterators · stdlib crypto/hpke + crypto/fips140 + ML-KEM
Swift	6.2+ (6.3 recommended)	Hummingbird 2 server · SwiftUI on iOS 26 / macOS 26
Kotlin	2.3.20+	Ktor 3 · Compose 1.9 BOM · Compose Multiplatform iOS 1.11+
C# / .NET	C# 14 + .NET 10 LTS	ASP.NET Core · WinUI 3 + MAUI · field keyword · primary constructors
C++	C++23 (clang 19+ / gcc 14+ / MSVC 2025+)	std::expected for Result · opportunistic C++26 via __cpp_lib_* gates
WASM/WASI	Preview 2 (wasi 0.2.x)	WIT 0.2 · Component Model · Preview 3 opt-in via --wasi-preview=3
Shell	POSIX sh + Bash 5+	POSIX.1-2024 baseline · Bash 5.2+ fallback · shellcheck-clean output · ADR-0032
Ruby	3.4+ (YJIT default)	RBS types · Rack/Sinatra HTTP · Fiber/Ractor · Data.define · ADR-0039

5 operating systems

macOS 14+ · arm64 + x86_64
Windows 11 · arm64 + x86_64
Linux: Ubuntu 24.04 / Debian 12 / Fedora 40+
iOS 26+ (--ios=18 opt-in)
Android API 34+

10 UI frameworks

SwiftUI · iOS 26 / macOS 26
Jetpack Compose · Compose 1.9 BOM (modules 1.11.1+)
Compose Multiplatform iOS · 1.11+ via SKIE
WinUI 3 · WinAppSDK 2.0+
.NET MAUI · via .NET 10 LTS
React · 19+ (18.3 floor)
Vue · 3.5+ (Composition API)
Svelte · 5+ (runes)
Solid · 1.9+ (createSignal)
Tauri · 2.11+ (native desktop via Rust)

10 cloud / edge runtimes

Vercel (Functions on Fluid Compute)
Cloudflare Workers (workerd, Hyperdrive, Workers AI, DO SQLite, Workers RPC, Agents SDK)
AWS Lambda (Node 22+, Python 3.13+, Go + Rust via provided.al2023)
GCP Cloud Run + Cloud Functions
Azure Functions (Flex Consumption + Durable Functions v3)
Bun 1.3+ as runtime
Deno 2.x as runtime
Fastly Compute@Edge
Netlify Edge Functions
Supabase Edge Functions

8 AI coding agents

Per ADR-0018, all eight are first-class authors. Claude Code is the mandatory final reviewer.

Claude Code

Primary + mandatory final reviewer (Opus 4.7 max-effort + 1M context)

OpenAI Codex

Alternative author

OpenCode (SST)

Alternative author

Cursor

Alternative author

Gemini CLI

Alternative author

Aider

Alternative author

Windsurf

Alternative author

Sourcegraph Amp

Alternative author

Install · pre-v1 placeholder

Reserved on 11 registries today

Every command below works now — but installs a v0.0.1 placeholder that prints a reservation notice. The real compiler ships at v1.0 Ludens.

We squatted these names defensively so no one can grab them. Pick your ecosystem on the full install page, or preview a few here.

npm install -g pseudo-lang

pip install pseudo-lang

brew tap alexfordlabs/pseudo && brew install pseudoc

mix archive.install hex pseudo_lang

See all 11 →

Conformance · MUST conform from v1.0 day 1

50+ standards across 9 categories

No "SHOULD conform." No "best-effort." No "v2 target." Every standard listed is fully conformant on v1.0 ship. Per-function @conforms makes claims auditable via pseudoc trace --conforms RFC 9106.

Foundational

BCP 14 · UAX #31 · UTF-8 · Unicode 17 · IEEE 754

Per-target

ECMA-262 · Edition 2024 · PEP 695/779 · Go 1.26 · Swift 6.2+ · Kotlin 2.3 · ECMA-334 (C# 14) · ISO/IEC 14882:2023 · WASM Core 3.0 · WASI Preview 2

Data formats

RFC 8259 (JSON) · TOML 1.0 · JSON Schema 2020-12 · OpenAPI 3.2 · CommonMark 0.31.2 · CycloneDX 1.7 · SPDX 3.0.1 · Source Map v3 · DWARF 5

Cryptography

RFC 9106 (argon2id) · FIPS 180-4/197/198-1/202/203/204 · RFC 8032 · RFC 7748 · RFC 9180 (HPKE) · RFC 8446 (TLS 1.3) · RFC 9562 (UUIDv7) · Paseto v4

HTTP / web

RFC 9110/9111/9112/9113/9114 · WHATWG Fetch + URL · CSP Level 3 · OAuth 2.0 · WebAuthn L3

Supply chain

SLSA v1.2 Build L3 · Sigstore · TUF 1.0.34 · in-toto v1.0 · SemVer 2.0 · Conventional Commits · Keep a Changelog · OSSF Scorecard ≥ 7.0

Community

Contributor Covenant 2.1 · DCO · MADR · CITATION 1.2.0 · agents.md · Anthropic Skill · MCP ≥ 2025-06-18 · LSP 3.18

Accessibility

WCAG 2.2 AA · WAI-ARIA 1.2

i18n

ECMA-402 · CLDR · BCP 47

Full list: CONFORMANCE.md · ~1,500-test conformance test suite at tests/conformance/v1.0/

Comments as language protocol · ADR-0027

Reasoning lives in the source, not in PR descriptions

Pseudo's distinctive thesis: comments aren't optional sugar — they're a typed AST. Every pub item MUST carry @intent and @authordoc comments, or it's a parse error. The next agent picking up the work inherits the reasoning trail intact.

Tier 1 — grammar-required

Parse error on every public item without these:

/// @intent  what this exists for
/// @author  model-name (YYYY-MM-DD)

Tier 2 — lint-required when applicable

Warnings via pseudoc check --comments:

/// @rationale  @invariant  @warning
/// @security   @measured   @origin
/// @conforms RFC 9106  (cited from CONFORMANCE.md)

22 frozen kinds across 3 buckets · v1.0 immutable · changes are v2.0 Sophia breaks

Doc kinds

@intent · @rationale · @invariant · @example · @param · @returns · @throws · @since · @deprecated · @stability · @security · @perf · @target:<name>

Provenance kinds

@author · @reviewer · @origin · @conforms

Inline kinds

@reason · @warning · @todo · @assumption · @measured

See: docs/style/comments.md · ADR-0027

Tools-first execution · ADR-0026

Four offline-first knowledge bases — one executable

Three read-only mirrors (~32 GB) of every language ecosystem, every standard, every framework — searchable instantly. Plus an executable tool inventory. AI agents bind to the canonical local tool, never npx, never curl | sh.

vendor/ · ~15 GB

Language source

9 ecosystems: rust, typescript, node, bun, deno, python, go, swift, kotlin, csharp, dotnet, cpp, llvm, wasm

standards/ · ~1.3 GB

Standards docs

12 bodies: ECMA, ISO, IETF RFC, W3C, WHATWG, Unicode, NIST, OWASP, supply-chain, data-formats, debug, markdown

platforms/ · ~16 GB

OS + framework

15 categories: apple, microsoft, android, linux, web/backend/mobile/desktop frameworks, all clouds, edge runtimes, AI platforms

tools/ · executable

Installed CLIs

192 dev tools across 15 categories. BY_PURPOSE.md= "I want to do X → use Y" canonical lookup.

Single-command refresh: ./scripts/refresh-all.sh (9 phases) · ADR-0026

RAG infrastructure · ADR-0028

Sub-100ms semantic search across 30+ GB of corpus

Local + remote Supabase + pgvector. Hybrid embedding model (bge-m3 default + voyage-code-3 opt-in). All 3 indexing tiers. Public read-only mirror lets contributors query the full corpus without indexing locally.

Hybrid retrieval

$ rag "argon2 parameters"

[1] 0.892  CONFORMANCE.md:55          RFC 9106 · argon2id...
[2] 0.871  standards/owasp/.../...md  m=46 MiB, t=1, p=1
[3] 0.834  spec §7.2:780              MUST use argon2id...
[4] 0.812  ADR-0027:155               @security on hash_password
[5] 0.798  rag/README.md:42           See: docs/style/comments.md

Public mirror — zero setup

# Anyone, no local indexing required
$ uv pip install pseudo-rag
$ rag --remote public "session token rotation"

# Vector cosine + BM25 + weighted rerank
# Across spec, ADRs, all 4 mirrors, RFCs, code

See: rag/README.md · ADR-0028

Release codenames

Psychology-themed milestones

v1.0 = Ludens. Apple/Google convention. Backwards-compatible v1.x.x; breaking changes wait for v2.0 Sophia.

v1.0

Ludens

Homo Ludens — language as play (Huizinga 1938)

v1.1

Freud — primal, foundational drive

v1.2

Ego

Freud — rational mediator

v1.3

Super Ego

Freud — norms, conformance, governance

v1.4

Libido

Freud — energy, drive

v1.5

Persona

Jung — the masks one wears

v1.6

Anima

Jung — soul, inner self

v1.7

Shadow

Jung — unconscious depths

v1.8

Self

Jung — wholeness, integration

v2.0

Sophia

wisdom — the redo, when v1 has run its course

.pseudo

Five things no other language gives you on day 1

11 native targets, one source

Comments are language protocol

Multi-agent first-class

Secure by default

No aspirational tiers

11 emit targets

5 operating systems

10 UI frameworks

10 cloud / edge runtimes

8 AI coding agents

Reserved on 11 registries today

50+ standards across 9 categories

Reasoning lives in the source, not in PR descriptions

Tier 1 — grammar-required

Tier 2 — lint-required when applicable

Four offline-first knowledge bases — one executable

Language source

Standards docs

OS + framework

Installed CLIs

Sub-100ms semantic search across 30+ GB of corpus

Hybrid retrieval

Public mirror — zero setup

Psychology-themed milestones

Be first when v1.0 Ludens ships