Today marks a major milestone for Nexa. With the release of v1.3.7, we've shipped 16 major features across 7 minor versions, backed by 1,500+ tests. Nexa has grown from a simple agent DSL into a full-featured programming language for autonomous systems.
The Journey from v1.1 to v1.3.7
Let's walk through what we've built, organized by priority tier:
🏆 P0: Core Differentiators
v1.1.0 — Intent-Driven Development (IDD): 104 tests. Introduced .nxintent files and the IAL (Intent Assertion Language) engine. Developers can now write natural-language specifications that are automatically verified against agent behavior. The @implements and @supports annotations create a traceable link between intent and implementation.
v1.2.0 — Design by Contract (DbC): 47 tests. Brought requires/ensures/invariant contracts to agent programming. Preconditions and postconditions are checked at runtime, with ContractViolation exceptions integrated across the entire runtime stack — from HTTP servers to database operations.
v1.3.0 — Agent-Native Tooling: 41 tests. Added nexa inspect, nexa validate, and nexa lint CLI commands. These aren't generic linters — they understand Nexa's agent semantics and can detect issues like missing protocol implementations or untyped agent outputs.
📦 P1: Essential Infrastructure
v1.3.1 — Gradual Type System: 79 tests. Three-mode type checking (STRICT/WARN/FORGIVING) with Int, String, Bool, List[T], Option[T], and Result[T,E] types. Type violations can be errors, warnings, or silently ignored — letting teams adopt typing at their own pace.
v1.3.2 — Error Propagation: 82 tests. The ? operator for Rust-style error propagation, plus otherwise handlers. NexaResult and NexaOption types with full monadic operations (map, unwrap, unwrap_or).
v1.3.3 — Background Job System: 73 tests. A full job DSL with priority queues, cron scheduling, exponential backoff, unique deduplication, and dead letter queues. Jobs can be defined declaratively in .nx files.
v1.3.4 — Built-In HTTP Server: 94 tests. A server DSL with static file serving, CORS/CSP configuration, route guards, middleware chains, hot reload, and semantic routing. Agents can be mounted as HTTP endpoints.
v1.3.5 — Database Integration: 84 tests. A db DSL supporting SQLite and PostgreSQL. Includes an Agent Memory API for storing and querying agent conversation history and knowledge.
🎯 P2: Advanced Capabilities
v1.3.6 — Auth, Concurrency, KV, Templates: 541 tests total. This was our biggest release:
- Auth & OAuth (84 tests): Three-layer authentication (API Key + JWT + OAuth PKCE), CSRF protection, session management
- Structured Concurrency (172 tests):
spawn/parallel/race/channelwith 18 API methods - KV Store (81 tests): SQLite-backed key-value store with TTL, atomic operations, and agent semantic queries
- Template System (209 tests): Jinja2-inspired
template"""..."""syntax with 30+ filters, partials, loops, and conditionals
✨ P3: Language Expressiveness
v1.3.7 — Pattern Matching, ADTs, and More: 543 tests. This release brought Nexa's expressiveness to a new level:
- Pattern Matching: 7 pattern types (literal, variable, wildcard, tuple, list, dict, variant) with destructuring
- Algebraic Data Types:
struct,enum,trait,impl— bringing Rust-like ADTs to agent programming - String Interpolation:
"Hello #{user.name}"syntax - Null Coalescing:
a ?? b ?? coperator - Defer Statements:
defer { cleanup() }for resource management - Pipeline Operator:
data |> transform |> output
By the Numbers
Total Features: 16
Total Tests: 1,500+
Latest Test Suite: 543 (v1.3.7)
Lines of Code: ~15,000 (Python) + ~8,000 (Rust AVM)
GitHub Stars: 1,200+
Contributors: 8
Release Cadence: ~2 weeks per minor versionWhat This Means
Nexa v1.3.7 is no longer a toy language. It has:
- A type system that catches errors before runtime
- A contract system that enforces behavioral guarantees
- An intent system that bridges natural language and code
- Production infrastructure: HTTP servers, databases, auth, job queues
- Language expressiveness: pattern matching, ADTs, templates, concurrency
But this is just the foundation. The real revolution is coming in v2.0 — where we introduce the Harness Native paradigm, making agent safety a compiler-enforced property rather than a runtime convention.
— Owen, April 2026