Kalandar

Kalandar is a natural-language capture parser for tasks, reminders, calendar events, and low-signal thoughts.

It ships three implementation paths behind one capture-preview contract:

• JavaScript for web and Node apps.
• Rust for Windows, Linux, and cross-platform native hosts.
• Swift/Core AI for Apple-native apps.

The Git repository is the distribution surface. It can be consumed as a JavaScript Git dependency or a Swift Package Manager dependency. npm registry and crates.io publication are disabled.

The repository includes parser code, runtime contracts, integration adapters, Swift source, and a validated resources/capture-parser ONNX suite for task, reminder, and calendar parsing. It does not include Core AI model artifacts, training data, checkpoints, local keys, or generated build output. A Core AI app supplies a validated artifact root separately.

Quick Start

Install from the private Git repository. Replace <git-ref> with a commit SHA, branch, or release tag that contains the version you want to consume:

npm install git+ssh://git@github.com/blinding-pixels/kalandar.git#<git-ref>
# or
pnpm add git+ssh://git@github.com/blinding-pixels/kalandar.git#<git-ref>

For Apple-native apps, add the repository directly through Swift Package Manager. See Use Swift and Core AI.

Parse a capture string:

import { parseCapturePreviewInput } from 'kalandar';

const preview = parseCapturePreviewInput('task: finish report tomorrow at 3pm #finance');

console.log(preview.kind);
console.log(preview.task);

Create a parser instance:

import { createCapturePreviewParser } from 'kalandar/capture';

const parser = await createCapturePreviewParser();
const preview = await parser.parse('reminder: water plants every other saturday at 9am');

Documentation

Start with the docs index.

• Getting started: install Kalandar and parse your first capture.
• Concepts: understand capture preview, implementations, artifacts, and inference.
• Tasks: use JavaScript, Rust, Core AI, or a self-hosted inference adapter.
• Tutorials: build an end-to-end capture preview flow.
• Reference: inspect the capture-preview-v1 request and response contract.
• Publishing: prepare private Git tags for JavaScript and SwiftPM consumers.

Implementation Paths

Path	Import or package	Best for
JavaScript	kalandar and kalandar/capture	Web apps, Node apps, and simple integrations.
Rust	rust/ source and kalandar-capture-preview CLI	Native desktop apps and cross-platform hosts.
Swift/Core AI	KalandarCoreAIRuntime SwiftPM product	Apple-native apps that use CoreAIParserSuite directly or expose a JavaScript bridge.

All three paths return the same capture-preview-v1 result shape.

Runtime Adapters

Kalandar exposes lower-level tensor runtime adapters from kalandar/runtime.

import { createRemoteInferenceAdapter } from 'kalandar/runtime';
import { createLocalOnnxInferenceAdapter } from 'kalandar/runtime/onnx';
import { createLocalCoreAiInferenceAdapter } from 'kalandar/runtime/coreai';

Install onnxruntime-node in Node apps that use local ONNX inference. The JavaScript Core AI adapter is a bridge for Apple hosts; native Swift apps can use CoreAIParserSuite directly.

The bundled ONNX resources are available at node_modules/kalandar/resources/capture-parser after a Git dependency install. Use a family directory such as task, reminder, or calendar with loadMlArtifact; use the parent capture-parser directory as the Rust CLI resource root.

Release Checks

From a source checkout:

npm run release:check

That command type-checks, builds, tests Swift and Rust, runs app-output parity, and performs an npm pack dry run. It does not validate separately distributed Core AI artifacts.

License

MIT