How Graftcode works under the hood

When to use this

You need one entry point for the full technical flow from public method to remote execution — without reading six separate concept articles.

Graftcode end-to-end flow — development time and runtime

What Graftcode Gateway is (and is not)

Graftcode Gateway (GG) is a native, dependency-free runtime host for Windows, Linux, and macOS. It loads application runtimes (.NET, JVM, Python, Node, …), discovers public interfaces, builds the UGM, and executes calls through embedded Hypertube and IIP.

GG is not a reverse proxy, API gateway, or traffic interceptor. It hosts code and connects runtimes — it does not sit between existing HTTP services.

Operational guides: Install GG, CLI reference. Full-duplex callbacks and multi-module hosting: Hypertube, Hosting multiple modules.

High-level flow

See the diagram at the top of this page. Step-by-step detail follows below.

Development time

Developer exposes public methods on modules (no REST required).
Graftcode Gateway loads runtimes, analyzes public interfaces, builds UGM (metadata only — no business logic leaves your environment).
UGM uploads to cloud when --projectKey / GSMU enabled, or serves locally for anonymous mode.
Package manager requests a Graft for calling technology (npm, NuGet, pip, …).
Graftcode Engine generates strongly typed client from UGM, caches, returns package.

See: How Grafts are generated, What goes to Graftcode Cloud.

Runtime

Caller invokes method on generated Graft (looks like local API).
Graft captures invocation intent — method name, args, types.
Intent encoded as IIP (Intention Invocation Protocol) binary message.
Hypertube on caller routes based on GraftConfig (in-memory, TCP, WebSocket).
GG on receiver loads target runtime, executes method, returns result or exception.
Result marshalled back to caller types; exceptions propagate natively.

See: Runtime call execution, Hypertube.

Key components

Component	Role
GG	Native runtime host; produces UGM; executes receiver-side calls
UGM	Language-agnostic contract (methods, types, signatures)
IIP	Binary protocol for programming intent
Hypertube	Runtime bridge; routing, sessions, full-duplex callbacks
Graft	Generated typed client in caller language
Graftcode Engine	On-demand graft generation from UGM
Virtual feed	Project-scoped registry of exposed module versions

Execution modes (same code)

Mode	GraftConfig	Where Hypertube runs call
In-memory	`host` unset	Same process
Local polyglot	`host` unset, multiple runtimes	Same process, different runtime
Remote	`host = tcp://…` or `ws://…`	Remote GG process

See: In-memory vs remote cookbook.

What is not sent to cloud

Business logic source or IL
Runtime data, secrets, or customer payloads
Only interface metadata (UGM) for graft generation

Deeper reading

Topic	Article
Gateway operations	Install GG, CLI reference
Caller/receiver	Caller and Receiver
Dev vs prod	Development-time vs production-time
Interface changes	What happens when interfaces change
Terms	Glossary

Backend pipeline (Grafting Agent)

The monorepo orchestration layer (GPMG → GPGE → GMA → GCG → GRPG → GSMU) implements cloud-side generation. That is internal platform architecture; operators typically interact via Portal and GG only.