This release adds full AI automation support: workflow orchestration with checkpointing, resource sandboxing, streaming/async, observability, and a cache layer. Plus universal improvements for binary serialization, tracing, sandbox results, and GenServer startup.
- Files:
lib/zixir/workflow.ex,lib/zixir/workflow/checkpoint.ex - Features: DAG execution, dependency management, automatic checkpointing, resume from failure, retry policies, dead letter queues
- Usage: Build fault-tolerant AI pipelines that recover automatically
- File:
lib/zixir/sandbox.ex - Features: Time limits, memory limits (e.g. "2GB"), CPU monitoring, automatic process termination, call depth limits
- Usage: Prevent runaway AI processes from consuming all resources
- File:
lib/zixir/stream.ex - Features: Async/await, parallel execution, lazy sequences, stream transformations (map, filter, batch), backpressure
- Usage: Handle streaming AI responses (LLMs) and parallel model inference
- File:
lib/zixir/observability.ex - Features: JSON logging, execution tracing with spans, Prometheus-compatible metrics, performance monitoring
- Usage: Monitor AI workflows without manual intervention
- File:
lib/zixir/cache.ex - Features: In-memory caching with TTL, disk persistence, database-like operations (insert/query/update), cache warming
- Usage: Store intermediate results and avoid redundant computation
| Improvement | Benefit |
|---|---|
| Binary serialization (instead of JSON) | Handles all Elixir data types (tuples, atoms, PIDs, etc.); more reliable for internal state persistence |
| Span struct initialization | Prevents KeyError in all tracing scenarios; works regardless of span lifecycle |
| Result wrapping in Sandbox | Follows Elixir {:ok, result} / {:error, reason} convention; consistent across all sandboxed operations |
| GenServer startup handling | Standard Elixir pattern for idempotent service startup; works in supervision trees and manual startup |
- Elixir 1.14+ / OTP 25+
- Zig 0.15+ (build-time; run
mix zig.getaftermix deps.get) - Python 3.8+ (optional) for ML/specialist calls
git clone https://github.com/Zixir-lang/Zixir.git
cd Zixir
git checkout v3.0.0
mix deps.get
mix zig.get
mix compile
mix testApache-2.0 — see LICENSE.