What You Inject Shapes What You Get
When you inject structured data into an LLM's context, the format directly affects comprehension. LLMs have been trained on billions of tabular examples - markdown tables, CSVs, log files. ISON leverages these familiar patterns to represent your data in a way LLMs naturally understand, using fewer tokens while preserving clarity. More efficient context means more room for the data that actually matters to your application.
Everything You Need for AI Workflows
From structured documents to streaming datasets to type-safe validation.
ISON is Built on Three Structures
- A collection of name/value pairs organized in blocks
- An ordered list of values organized in tables
- A reference mechanism for expressing relationships
These are universal patterns for structured data. Virtually all AI systems need them in one form or another. It makes sense that a data format designed for LLMs should represent these structures in the most token-efficient way possible.
JSON vs ISON
Same data, but ISON uses familiar tabular patterns instead of JSON's verbose syntax. Notice how JSON repeats key names for every row and adds quotes, colons, and braces throughout.
JSON (~280 tokens)
{
"users": [
{
"id": 1,
"name": "Alice",
"email": "alice@example.com",
"active": true
},
{
"id": 2,
"name": "Bob",
"email": "bob@example.com",
"active": false
}
]
}ISON (~140 tokens)
~70% more context capacitytable.users
id name email active
1 Alice alice@example.com true
2 Bob bob@example.com falseUse Your Favorite Language
Production-ready parsers and validation libraries for major programming languages.
Common Context Injection Scenarios
Anywhere you inject structured data into LLM context, ISON can help you fit more relevant information.
RAG Pipelines
Include more retrieved documents in the same context window. More relevant context helps LLMs provide better, more grounded answers.
Agentic Workflows
MCP tools, function calls, and agent state all inject data into context. Efficient representation leaves room for reasoning.
Database Results
Query results are inherently tabular. ISON represents them naturally without the overhead of nested JSON structures.
System Prompts
User profiles, permissions, and domain knowledge can be compactly represented, leaving more room for instructions.
Streaming Data (ISONL)
For logs, events, and training data, ISONL provides a line-based format that's efficient for both storage and context injection.
Knowledge Graphs
Native reference syntax allows you to represent relationships between entities without verbose nested structures.