# PrismLang

One canonical identity. Human page: https://www.insightits.com/products/prismlang.html

| Field | Value |
|-------|-------|
| Slug | `prismlang` |
| Status | published |
| Family | supporting |
| Infra category | — |
| Record type | product |
| Parent | — |
| JSON | https://www.insightits.com/catalog/prismlang.json |
| Markdown | https://www.insightits.com/catalog/prismlang.md |
| GitHub | https://github.com/insightitsGit/prismlang |
| PyPI | https://pypi.org/project/prismlang/ |
| Install | `pip install prismlang. Apache 2.0. 57–62% token reduction by domain across healthcare, finance, and trade.` |
| Version | — |

## What it is

Middleware layer for LangGraph — not a replacement. Add @prism_node on existing graphs for leaner inter-agent state, audit trail, and tenant isolation. Zero LLM dependency for routing.

## Problem

Inter-agent LangGraph state is token-heavy. Teams that already run LangGraph need compression and deterministic routing without another LLM hop.

## Alternatives

- **Plain LangGraph state passing** — PrismLang sits on top of existing graphs. Domain token-reduction figures are product-authored (healthcare / finance / trade), not a named-vendor bake-off page.

## Features

- **57–62% Token Reduction** — Compress inter-agent state transport—not context windows—with proven 57–62% savings across domains.
- **Full Audit Trail** — Every envelope carries an immutable rule_chain tracing taxonomy → projection → JL reduction.
- **Tenant Isolation** — SHA-256-seeded JL matrices per tenant—cross-tenant cosine similarity stays below 0.20.

## Architecture

Tenant-isolated 64-d vectors, ONNX CPU encoding. Every envelope carries an immutable rule_chain (taxonomy → projection → JL reduction). SHA-256-seeded JL matrices per tenant.

## Use cases

- Existing LangGraph graphs that hop state between agents
- Healthcare, finance, and trade graphs where inter-agent tokens dominate

## Benchmarks

Product-authored domain token-reduction (3 turns): Healthcare −62.1% / Finance −57.0% / Trade −58.6% prompt tokens; encode ~31–35 ms CPU; LLM latency unchanged.

### Disclosures

- Vendor-authored domain benchmarks, not a named-vendor bake-off.



## GitHub

https://github.com/insightitsGit/prismlang

## PyPI

https://pypi.org/project/prismlang/

## Documentation

- [GitHub README](https://github.com/insightitsGit/prismlang#readme)
- [Quick start](https://github.com/insightitsGit/prismlang#quick-start)
- [issues](https://github.com/insightitsGit/prismlang/issues)
- [docs](https://github.com/insightitsGit/prismlang/tree/master/docs)
- [paper](https://www.insightits.com/prismlang/paper)

## Is not

- A LangGraph replacement
- Needed on the ChorusGraph native path without LangGraph product graphs

## Related products

- [python-libs](./python-libs.md)
- [chorusgraph](./chorusgraph.md)