A DAG workflow engine with validated dependencies, concurrent async execution, plugin tasks, resumable state, and run-level observability.
Multi-step jobs need more than a task list: dependencies must be valid, failures must propagate clearly, and interrupted work must remain inspectable.
Built the graph validator, execution manager, plugin protocol, persistent task state, FastAPI control plane, YAML examples, CLI, JSONL logging, tests, CI, and Docker runtime.
Concurrency, retries, timeouts, cancellation, dependency failure, and resume behavior had to produce a coherent, persistent task history.
SQLite is the default local state store.
Built-in plugins are deliberately small demonstration tasks.
The public project does not claim a hosted production service.
The engine validates the graph before execution, schedules topological levels under a bounded semaphore, resolves typed task plugins, persists every transition, and exposes trigger, inspect, cancel, resume, and graph endpoints.
Cycle and missing-dependency validation
Bounded concurrent execution
Retries and per-task timeout limits
Cancellation tokens and dependency-aware skipping
Persistent workflow and task state
Per-run JSONL operational logs
A compact, installable workflow engine that demonstrates graph reasoning, stateful async execution, API design, packaging, and strict CI quality gates.
Tayyab can design orchestration behavior where state transitions and failure semantics are first-class concerns.
Validate the complete graph before any side effect starts.
Persist state around task execution so interruption is observable.
Plugin boundaries keep the scheduler independent from task implementation.