Introduction

1.1 Project Overview

1.1.1 What is AmritaSense?

AmritaSense is a next-generation, general-purpose workflow and event stream engine. It abandons traditional graph-based design and instead adopts the execution model of a low-level computer—an instruction set, a pointer vector, and a program counter—to orchestrate and control workflows.

Unlike systems that require you to explicitly define nodes and edges and then rely on a scheduler to interpret the graph structure, AmritaSense provides an intuitive composition model. You simply combine nodes and control structures as if writing ordinary Python code; the framework compiles them into a linearized instruction sequence and a lightweight “virtual machine” executes them step by step. This design gives AmritaSense unparalleled expressive power and runtime control precision while keeping overhead extremely low.

Furthermore, AmritaSense includes a complete event and dependency injection subsystem with no external dependencies. Workflow nodes can trigger custom events, and event handlers share the same dependency injection mechanism as nodes, unifying orchestration and reaction inside a single runtime.

1.1.2 Why AmritaSense?

We believe that workflows should be designed for the work, not limited by the flow. “Flow” is merely a presentation; it should never become a shackle when you design logic.

• Natively Turing-complete – You can implement complete, arbitrary control logic without defining complex boundary conditions inside your program. AmritaSense natively supports conditionals (IF/ELIF/ELSE), loops (WHILE/DO‑WHILE), unconditional jumps (GOTO), subroutine calls (CALL), and exception handling (TRY/CATCH)—no external graph engine or state machine required.
• Virtual-machine addressing model – AmritaSense uses a classic computer-style addressing and execution model (PointerVector + call stack). All high-level control flow is expanded into uniform pointer instructions at compile time. Only integer operations and function calls remain at runtime; scheduling overhead is nearly zero.
• Built for AI agents and complex business logic – Whether it is tool-calling loops, nested sub‑workflows, pausing for user input, or exception recovery and rollback, AmritaSense can express it directly and execute it with extreme efficiency.

1.1.3 Core Value Proposition

• Low-level computer execution model – Replaces topological graphs with a pointer vector and call stack; minimal runtime, far outperforming traditional graph engines.
• Natively Turing-complete – IF, WHILE, DO‑WHILE, GOTO, CALL, TRY, INTERRUPT and more are all natively supported—no simulation.
• Focus on flow logic – Say goodbye to complicated edge definitions and state dictionaries; concentrate on the business process you actually care about.
• Declarative dependency injection – Nodes and event handlers declare required parameters through function signatures; the framework automatically performs keyword matching and type resolution.
• Async-first, suspendable & resumable – Native async/await support, built-in full-duplex streaming primitive (SuspendObjectStream), allowing precise suspension and resumption between nodes.
• Interpreter tree & subgraph isolation – Fork sub-interpreters for parallel execution; manage entire interpreter trees with wait_all / terminate_all; use FUN_BLOCK for isolated sub-workflow calls with independent middleware and error boundaries.
• Self-contained runtime – Built-in logging system, event bus, and dependency injection; zero mandatory external dependencies; embeddable into any Python project.