HomeDocsJavaScript / TypeScript
npm · brevit · v1.0.2View on npm

JavaScript / TypeScript SDK

Brevit for JavaScript / TypeScript — compress JSON, text, and binary data before sending to LLMs. Reduce token costs by 40–60% with zero configuration.

Getting Started

Brevit is a drop-in library that transforms your data into a token-efficient format before it reaches your LLM. It works on JSON structures, plain text, and binary image data — and it's completely lossless by default.

Tip:Not sure which SDK to use? All three have identical output. Pick the one that matches your stack.

Installation

Terminal
$ npm install brevit

Requires Node.js 18+ or a modern browser. ESM and CommonJS both supported.

Quick Start
import { BrevitClient, BrevitConfig } from 'brevit';const brevit = new BrevitClient();// Auto mode — Brevit picks the best strategyconst result = await brevit.brevity({  user: { name: 'Jane', email: 'jane@example.com' },  orders: [    { id: 'ORD-001', status: 'SHIPPED', total: 79.98 },    { id: 'ORD-002', status: 'PENDING', total: 29.99 }  ]});console.log(result);// user.name:Jane// user.email:jane@example.com// orders[2]{id,status,total}:// ORD-001,SHIPPED,79.98// ORD-002,PENDING,29.99

Configuration

Pass a config object to BrevitConfig to customize optimization behavior. All fields have sensible defaults — zero config is required for most use cases.

Configuration
import { BrevitClient, BrevitConfig, JsonOptimizationMode, TextOptimizationMode } from 'brevit';const config = new BrevitConfig({  // JSON optimization strategy  jsonMode: JsonOptimizationMode.Flatten,  "color:#546e7a">// Flatten | ToYaml | Filter | None  // Text optimization strategy  textMode: TextOptimizationMode.Clean,    "color:#546e7a">// None | Clean | SummarizeFast | SummarizeHighQuality  // For Filter mode: paths to keep  jsonPathsToKeep: [],  // Minimum characters before text optimization triggers  longTextThreshold: 500,  // Enable @alias=prefix abbreviations (v0.1.2+)  enableAbbreviations: true,  // Min times a prefix must repeat before getting an alias  abbreviationThreshold: 2,});const brevit = new BrevitClient(config);
OptionDefaultDescription
jsonModeFlattenHow to optimize JSON. Flatten produces dot-notation key-value pairs.
textModeCleanHow to handle long text strings in JSON values.
imageModeOcrImage optimization — OCR extracts text from images.
enableAbbreviationstrueAuto-abbreviate repeated key prefixes with @alias syntax.
abbreviationThreshold2Min times a prefix must repeat before it gets an alias.
longTextThreshold500Char count threshold before text optimization triggers.

JSON Optimization

Brevit's JSON optimizer uses a smart detection pipeline: nested objects become dot-notation paths, uniform object arrays become compact tabular format, and primitive arrays become comma-separated.

Nested Objects

user: { name: "J" }
↓
user.name:J
~38% savings

Primitive Arrays

["a", "b", "c"]
↓
arr[3]:a,b,c
~70% savings

Object Arrays

[{ id:1 }, { id:2 }]
↓
arr[2]{id}:
1
2
~60% savings
JSON Optimization Examples
const brevit = new BrevitClient();// Nested object → dot notationconst flat = await brevit.optimize({  user: {    id: "u-123",    name: "Javian",    contact: { email: "hi@example.com", phone: "+1-555-0123" }  }});// user.id:u-123// user.name:Javian// user.contact.email:hi@example.com// user.contact.phone:+1-555-0123// Primitive array → comma-separatedconst arr = await brevit.optimize({ tags: ["json", "llm", "api"] });// tags[3]:json,llm,api// Uniform object array → tabular formatconst table = await brevit.optimize({  products: [    { sku: "A1", qty: 3, price: 9.99 },    { sku: "B2", qty: 1, price: 24.99 },  ]});// products[2]{sku,qty,price}:// A1,3,9.99// B2,1,24.99

Abbreviation Engine

When the same key prefix appears ≥ abbreviationThreshold times, Brevit creates a short alias and replaces the prefix throughout — saving an additional 10–25% on top of flattening.

Note:Abbreviations are only created when the token savings exceed the cost of writing the definition. If abbreviating wouldn't save tokens net, the prefix is left as-is.
Abbreviation Examples
const config = new BrevitConfig({ enableAbbreviations: true, abbreviationThreshold: 2 });const brevit = new BrevitClient(config);const result = await brevit.optimize({  user: { name: "John", email: "john@example.com", role: "admin" },  order: { id: "o-456", status: "SHIPPED", total: 99.99 }});// Output with abbreviations:// @u=user// @o=order// @u.name:John// @u.email:john@example.com// @u.role:admin// @o.id:o-456// @o.status:SHIPPED// @o.total:99.99// Without abbreviations (disabled):const noAbbrevConfig = new BrevitConfig({ enableAbbreviations: false });const result2 = await new BrevitClient(noAbbrevConfig).optimize({ ... });// user.name:John// user.email:john@example.com// ...

Alias Generation Strategy

  1. 1. First letter of root segment: user → u
  2. 2. First letter of each dot-segment: order.items → oi
  3. 3. Counter-based fallback: a, b, c … z, aa, ab…

Text Optimization

Brevit uses a deterministic TextRank / PageRank-style algorithm over sentence similarity graphs. No LLM is required — it's purely extractive and always produces the same output for the same input.

AUTO Mode

With autoThresholdMultiplier = 0.0, all sentences pass the threshold → lossless by default.

RATIO Mode

Pass a ratio 0.0–1.0 to keep the top N% of sentences ranked by TextRank graph score.

Text Optimization Examples
const brevit = new BrevitClient();// compressText() — AUTO mode (lossless by default)// Runs TextRank analysis; with autoThresholdMultiplier=0, all sentences are kept.const compressed = await brevit.compressText(longDocument);// optimizeText() — RATIO mode (0.0 to 1.0)// Keeps the top-ranked N% of sentences by TextRank scoreconst summary = await brevit.optimizeText(longDocument, 0.5);  "color:#546e7a">// keeps 50%// Through optimize() with ratioconst result = await brevit.optimize(longDocument, 0.3);  "color:#546e7a">// keeps 30%// Through brevity() — auto-detects text and routes to compressTextconst auto = await brevit.brevity("This is a long article about AI...");

Image Optimization

Warning:Image OCR is a stub in the current release. The interface is defined and ready to plug in Azure AI Vision, Tesseract, or any OCR provider via a custom optimizer.

Pass a Uint8Array (JS),bytes (Python), orbyte[] (.NET) to the client and Brevit will route it through the configured image optimizer.

Custom Strategies

Extend Brevit with your own optimization logic. Register a strategy with an analyzer that scores how well it fits the data, and an optimizer that performs the transformation.

Custom Strategy Registration
const brevit = new BrevitClient();// Register a custom strategybrevit.registerStrategy(  'my-strategy',  // Analyzer: returns score 0-100 for how well this strategy fits the data  (data) => ({    score: typeof data === 'object' && data.type === 'invoice' ? 95 : 0,    reason: 'Invoice-specific optimization'  }),  // Optimizer: the actual transformation  async (data) => {    return `INV#${data.id} ${data.vendor} $${data.total}`;  });

API Reference

Core methods available on BrevitClient. All async methods return Promise<string> (JS/Python) or Task<string> (.NET).

MethodPurpose
brevity(data)Auto-selects best strategy based on data analysis. Recommended entry point.
optimize(data, ratio?)Main pipeline: routes by type (JSON/text/image/POCO). Optional ratio for text.
compressText(text)Explicit TextRank AUTO mode — lossless by default.
optimizeText(text, ratio)Explicit TextRank RATIO mode — keeps top N% of sentences.
registerStrategy(name, analyzer, optimizer)Register a custom optimization strategy for use in brevity().
View full API reference

Real-World Examples

E-Commerce Orders

Compress customer orders and product data for LLM analysis.

Document Processing

Summarize legal documents and contracts before LLM review.

RAG Pipelines

Optimize retrieval chunks before injecting into LLM context.

Customer Support

Compress ticket history and chat logs for agent context.

Try the Playground