Developer manual

Muxiva serves two extension paths. Application teams often already own an Agent and need to connect it to real-time audio, interruption, multimodal data, and observability. Infrastructure teams build ASR, TTS, RTC, and media Nodes.

Both enter the same typed Graph as project Node Packages, but their ownership boundaries differ:

Goal	Start here
Deploy an existing Agent into Muxiva	Agent integration
Try the file-capable reference Agent	Pi coding Agent
Build a Python algorithm or application Node	Python
Build a TypeScript Agent or application Node	TypeScript
Build a high-performance Rust Node	Rust
Wrap an RTC, media, or vendor C++ SDK	C++

Project layout

my-agent/
├── graph.json                         # Nodes, Ports, Edges, configuration
├── package.json                       # Agent dependencies and lock file
├── .env                               # local credentials, Git ignored
└── .muxiva/
    ├── nodes/my_agent/
    │   ├── muxiva.node.json           # Node contract
    │   └── node.ts                    # thin Agent → Muxiva adapter
    ├── agents/my-agent/               # pinned application-owned Agent repo
    └── workspaces/my-agent/            # file authority granted to the Agent

graph.json never contains executable source. A muxiva.node/v1 Manifest declares discovery and validation metadata. The Agent remains independently versioned, tested, reviewed, and released by its application team.

Node Manifest

Field	Purpose
`package_id`	filesystem-safe project identity
`display_name`	Studio Palette label
`node_type`	stable Factory type
`language`	`rust`, `cpp`, `python`, or `typescript`
`factory_version`	exact Factory or adapter version
`kind`	`source`, `transform`, or `sink`
`category`	transport, algorithm, media, control, or utility
`capability`	stable searchable capability
`entrypoint`	language implementation entrypoint
`ports`	names, direction, Frame types, and semantic schemas
`config_schema`	JSON Schema for Node configuration

Studio-first workflow

Ordinary Nodes can be authored directly in Studio:

open Studio for the project Graph;
choose Create Node;
define the language, role, typed Ports, and configuration schema;
edit code and choose Save & Register;
drag the Node from the Palette and connect compatible Ports;
Validate, Run, and inspect calls, queues, messages, and media in Observe.

Do not paste an entire existing Agent into Studio. Keep the Agent in its own repository and store only a stable adapter under .muxiva/nodes/. Follow the complete Agent integration SOP.