TokenSlim
v2.0
v2.0 · Open-source · MIT

Compress LLM tokens
by 40–70%

Type preservation, value dictionaries, and 7+ integrations. One import change. Zero model-side changes.

Try the playgroundView on GitHub →
40–70%
Token savings
98.4%
LLM accuracy
1 line
Code change
7+
Integrations
input.json351 chars
[
  {
    "employeeId": "EMP-001",
    "firstName": "Alice",
    "department": "Engineering",
    "salary": 125000
  },
  {
    "employeeId": "EMP-002",
    "firstName": "Bob",
    "department": "Marketing",
    "salary": 95000
  },
  {
    "employeeId": "EMP-003",
    "firstName": "Carol",
    "department": "Engineering",
    "salary": 130000
  }
]
output.slim187 chars · 47% smaller
@keys{employeeId:_0,firstName:_1,department:_2}

[3]{department,employeeId,firstName,salary}:
Engineering,EMP-001,Alice,125000
Marketing,EMP-002,Bob,95000
Engineering,EMP-003,Carol,130000

Live Playground

Paste JSON or pick a sample to see SLIM compression in real-time. Toggle @types, @dict, and deterministic mode for v2.0 features.

Samples:
Delimiter:
Input
Output
SLIM output appears here...

Cost Calculator

Estimate how much you'll save per month with TokenSlim.

10%70%
Monthly savings
$50.00
Yearly savings
$600.00
Monthly cost (before)
$125.00
Monthly cost (after)
$75.00

Benchmark Suite

SLIM vs JSON vs TOON vs YAML vs CSV across real-world datasets.

DatasetShapeJSONSLIMTOONYAMLCSV
Medium User List (50)8 cols × 50 rows2065baseline54%1083 tok54%1081 tok15%2015 tok56%1029 tok
Nested Config3 levels deep176baseline0%207 tok4%194 tok0%206 tokN/A
API Response (50 orders)Mixed structure3658baseline61%1645 tok57%1784 tok7%3881 tokN/A
Log Entries (40)7 cols × 40 rows1696baseline58%813 tok56%853 tok15%1646 tok58%812 tok
Product Catalog (30)7 cols × 30 rows889baseline61%392 tok61%400 tok13%881 tok62%383 tok

Token counts are estimates. CSV lacks lossless round-trip for nested data. SLIM provides the best balance of savings and round-trip fidelity.

Run locally:npx @tokenslim/benchmarkUse --json for CI, --format / --dataset for filtering

Integrate in one minute

Change one import. Keep your existing code.

npm install @tokenslim/openai openai
BEFORE
import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});

const response = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [
    { role: 'user', content: JSON.stringify(data) },
  ],
});
AFTER (TOKENSLIM)
import { TokenSlimOpenAI as OpenAI } from '@tokenslim/openai';

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});

const response = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [
    { role: 'user', content: JSON.stringify(data) },
  ],
});

That's it. All SDK features, types, and error handling work identically.