AINL + Temporal: best of both worlds

Temporal gives you durable execution, retries, and worker infrastructure. AINL gives you a compact authoring surface, strict validation, and canonical graph IR that compiles once and runs many times without the LLM re-deriving control flow on every run.

You do not have to pick one: from v1.2.5, validate can emit Temporal-shaped Python modules from the same IR that powers the core runtime (current release: v1.3.3).

How it fits

1. Author operational workflows in AINL (monitors, token trackers, SLA checks, bridge-driven workers).

2. Compile with --strict so reachability and adapter contracts are checked before production.

3. Emit when you need Temporal’s worker model:

   python3 scripts/validate_ainl.py --strict path/workflow.ainl --emit temporal -o ./out/prefix
   

   See docs/hybrid_temporal.md and examples/hybrid/temporal_durable_ainl/ for the intended layout.

4. Run activities/workflows with Temporal’s test or production environment; keep .ainl as the single source of truth and regenerate emit when the graph changes.

What AINL is not replacing

• Temporal’s server, namespaces, visibility, and worker scaling — still Temporal’s job.
• Long-lived human-in-the-loop conversational state — out of scope for strict operational graphs (use adapters + explicit frames).

Tests in this repo

• Optional temporalio integration tests emit and exercise emitted code (see tests/test_hybrid_emit_integration.py).

Related

• Hybrid guide — full hybrid story (LangGraph + Temporal + S hybrid hints).
• Runner discovery: GET /capabilities/temporal on the HTTP runner.
• From LangGraph to AINL — onboarding from graph-first Python frameworks.