Implementations

Implementations Type-safe

Development flow

Define a tool in the nara webapp (name, parameters schema, result schema, execution type EDGE or DESKTOP).
Run edge-connector generate to scaffold language-specific stubs.
Fill in the business logic in the generated files.
Use edge-connector test to execute the tool locally.
Package and upload your implementations when you are ready.

Typical structure:

custom-implementations/
├─ package.json
├─ tsconfig.json
├─ index.ts              # exports all tools
└─ tools/
 └─ my_tool.ts         # generated stub for 'my_tool'

The generated tool stub includes:

a strongly typed input parameter object validated against the tool schema,
a typed result shape,
a context object with user, org, and deployment metadata.

Implement your logic in the generated functions; keep the signatures intact so the CLI and runtime can type-check them.

For Python implementations, the generator creates:

custom-implementations/
├─ requirements.txt
├─ main.py
└─ tools/
 └─ my_tool.py

The Python stubs follow the same contract as the TypeScript ones: a strongly-typed args object, a context object, and a structured result matching the tool’s result schema.

Context object basics

Tool implementations receive a context object that typically includes:

user identity (for example userId)
organization ID
deployment metadata (hostname, platform)

Use this to:

enforce your own authorization checks,
route requests to the correct downstream systems,
add audit logging to your own observability stack.